cBioPortalChat test sets

Each questions file is a separate test set with its own purpose. A run asks every question in one set; its results are on the results page, one table per set.

Set	Questions	Purpose	Runs
Main benchmark `input/questions.yaml`	146	The headline score. Each question stands alone: a count, frequency or list from the data, an analysis (comparison, survival, co-occurrence), a request for a cBioPortal link, or something the agent should decline or redirect. Use it to compare prompts, models and runners.	7
Multi-turn follow-ups `input/questions-multiturn.yaml`	14	About 40% of real messages are follow-ups that need the earlier conversation. Each question is the user's next message after scripted turns: swapping a gene, study or cohort, breaking results down, accepting an offered option, answering a clarifying question, pushing back on a correct number, asking for significance or citations, and follow-ups in other languages.	3
Subset questions `input/questions-subset.yaml`	26	The main set leans on per-study statistics (top genes in a study, a gene's frequency in a study), which precomputed tables and per-study tools answer directly. Here the cohort is a clinical subset of a study (sample type, sex, smoking, stage, MSI, receptor status, OncoTree code) or a pooled set of studies, so the agent must write the SQL and must not answer with the whole-study number. Covers subset frequencies, pooled studies, subset comparisons, co-occurrence, top genes, patient vs sample level, clinical counts in mutation-defined groups and multi-turn narrowing.	1

Questions

Main benchmark

The headline score. Each question stands alone: a count, frequency or list from the data, an analysis (comparison, survival, co-occurrence), a request for a cBioPortal link, or something the agent should decline or redirect. Use it to compare prompts, models and runners.

Questions: input/questions.yaml · 7 runs

62 Data26 Navigation46 Analysis12 Out of scope

11 Study discovery13 Cohort & clinical counts36 Alteration frequency16 Variants & hotspots8 Co-occurrence & exclusivity20 Expression & multi-omics14 Survival & outcomes6 Treatment10 Patient & sample lookup12 Out of scope

uv run cbioportal-mcp-qa run --questions-file input/questions.yaml

#1 DataStudy discovery · All StudiesHow many studies are in cBioPortal?

Reference answer (checked 2026-09-23): 548
Expected links: https://www.cbioportal.org/datasets
https://www.cbioportal.org
Notes: The answer should match the total number of studies shown on the home page and the DataSets page.

#2 DataStudy discovery · All StudiesHow many glioblastoma studies are in cBioPortal?

Reference answer (checked 2026-09-23): 8
Expected links: https://www.cbioportal.org/datasets
https://www.cbioportal.org
Notes: The answer should match that shown on the home page. In Select Studies for Visualization & Analysis, enter: glioblastoma in the search box. You should get the same number via the DataSets page.

#3 DataCohort & clinical counts · msk_chord_2024How many patients and samples are in the MSK-CHORD Study?

Reference answer (checked 2026-09-23): 24,950 Patients and 25,040 samples
Expected links: https://www.cbioportal.org/study/summary?id=msk_chord_2024

#4 DataCohort & clinical counts · msk_chord_2024How many primary samples are in the MSK-CHORD Study?

Reference answer (checked 2026-09-23): 15,928 primary samples
Expected links: https://www.cbioportal.org/study/summary?id=msk_chord_2024
Notes: Expected link view: Hover over “Sample Type” chart to see sample type numbers.

#5 DataTreatment · msk_chord_2024What treatment did most patients receive in the MSK-CHORD Study?

Reference answer (checked 2026-09-23): Fluorouracil
Expected links: https://www.cbioportal.org/study/summary?id=msk_chord_2024

#6 DataAlteration frequency · os_target_gdcWhat are the top 5 most frequently mutated genes in the Osteosarcoma study from TARGET?

Reference answer (checked 2026-09-23): TP53 22.4%, MUC16 11.2%, TTN 11.2%, ATRX 7.7%, DNAH9 7.0%
Expected links: https://www.cbioportal.org/study/summary?id=os_target_gdc

#7 DataAlteration frequency · os_target_gdcWhat are the top 5 most frequently copy number altered genes in the Osteosarcoma study from TARGET?

Reference answer (checked 2026-09-23): LINC00901 (deep deletion) 39.5% (32/81 CNA-profiled samples); RN7SL442P (amplification) 38.3% (31/81); then a tie at 37.0% (30/81) among 17p11.2 amplicon genes such as DRG2, FLII, LLGL1, UBB, NCOR1 and MYO15A.
Expected links: https://www.cbioportal.org/study/summary?id=os_target_gdc
Notes: Ties: after the top two, any genes from the 17p11.2 amplicon at 37.0% are acceptable. Counts are AMP + HOMDEL over the 81 CNA-profiled samples.

#8 DataAlteration frequency · os_target_gdcWhat are the top 5 most frequently altered genes in a structural variant in the Osteosarcoma study from TARGET?

Reference answer (checked 2026-09-23): The TARGET osteosarcoma study (os_target_gdc) has no structural variant data, so no genes can be ranked by SV frequency.
Notes: A correct answer must say the study has no structural variant profile; must not list fusions or SV genes for this study.

#9 DataAlteration frequency · os_target_gdc"What are the top 5 frequently altered genes in the Osteosarcoma study from TARGET for mutations, copy numbers and SVs combined?"

Reference answer (checked 2026-09-23): TP53 24.1% (38/158 samples profiled for mutations or CNA), LINC00901 20.3% (32/158), then CSMD3, MYO15A, TRIM16 and RN7SL442P tied at 19.6% (31/158). The study has no SV data, so this combines mutations and CNAs only.
Expected links: https://www.cbioportal.org/study/summary?id=os_target_gdc
Notes: Denominator choice changes percentages (158 = union of mutation- and CNA-profiled samples); ranking with TP53 first is the key point. Must note there is no SV data.

#10 DataCohort & clinical counts · os_target_gdcWhat is the median age at diagnosis for osteosarcoma patients in the TARGET study?

Reference answer (checked 2026-09-23): About 15 years: median 15.2 years at diagnosis (from DAYS_TO_BIRTH, n=293 patients; range 3.6–87). The study's AGE attribute is floored at 18 for 241 of 293 patients, so a median of 18 from AGE is misleading.
Expected links: https://www.cbioportal.org/study/summary?id=os_target_gdc
Notes: Answers of 18 based on the AGE attribute are partially correct only if they flag that AGE appears capped at 18.

#11 DataAlteration frequency · msk_chord_2024What are the top 5 most frequently mutated genes in the MSK-CHORD Study?

Reference answer (checked 2026-09-23): TP53 52.4%, KRAS 28.5%, APC 19.1%, PIK3CA 14.8%, EGFR 8.6%

#12 DataCohort & clinical counts · msk_chord_2024What is the most common cancer type in the MSK-CHORD Study based on sample count?

Reference answer (checked 2026-09-23): Non-Small Cell Lung Cancer: 7,809 of 25,040 samples (31.2%), followed by colorectal (5,543) and breast (5,368).
Expected links: https://www.cbioportal.org/study/summary?id=msk_chord_2024

#13 DataAlteration frequency · msk_chord_2024What percentage of patients in the MSK-CHORD Study have at least one TP53 mutation?

Reference answer (checked 2026-09-23): 52.5% of patients (13,105 of 24,950) have at least one TP53 mutation (52.4% of samples: 13,124 of 25,040).
Expected links: https://www.cbioportal.org/results/mutations?cancer_study_list=msk_chord_2024&case_set_id=msk_chord_2024_sequenced&gene_list=TP53

#14 AnalysisCohort & clinical counts · All StudiesWhich cancer type has the highest average tumor mutational burden across all studies?

Reference answer (checked 2026-09-24): Depends on the statistic, in the TCGA PanCancer Atlas studies (TMB_NONSYNONYMOUS): highest mean is uterine endometrial carcinoma (UCEC, 35.66, driven by hypermutators), then melanoma (SKCM, 25.91); highest median is melanoma (SKCM, 14.88), then LUSC 7.67, LUAD 6.68, BLCA 5.77.
Notes: A correct answer must: pick one consistently processed cohort (e.g. the TCGA PanCancer Atlas studies, *_tcga_pan_can_atlas_2018) rather than averaging TMB across studies with different panels/pipelines, say which TMB attribute and which statistic (mean or median) it used, and name the top cancer type for that statistic. UCEC (mean) and SKCM (median) are both correct when the statistic is stated. Must not: average TMB across heterogeneous studies without caveats, present a single number without the cohort, or call melanoma the highest mean without qualification.

#15 DataStudy discovery · All StudiesHow many total studies contain mutation data in the cBioPortal database?

Reference answer (checked 2026-09-23): 542 studies

#16 DataCohort & clinical counts · msk_chord_2024How many unique patients have both primary and metastatic samples in the MSK-CHORD Study?

Reference answer (checked 2026-09-23): 26 patients

#17 DataCohort & clinical counts · msk_chord_2024What are the top 5 most common primary diagnosis sites in the MSK-CHORD Study?

Reference answer (checked 2026-09-23): By sample count: Lung 7,773; Breast 5,358; Prostate 3,208; Pancreas 3,104; Colon 2,516 (of 25,040 samples, PRIMARY_SITE).
Expected links: https://www.cbioportal.org/study/summary?id=msk_chord_2024
Notes: Grouping colon + rectum + sigmoid colon as colorectal is also acceptable if stated.

#18 DataTreatment · msk_chord_2024What are the most frequently administered systemic therapy regimens for lung cancer patients in the MSK-CHORD Study?

Reference answer (checked 2026-09-24): Among Non-Small Cell Lung Cancer patients (CANCER_TYPE, 7,809 patients), grouping agents given on the same day into regimens and excluding Investigational, Prior Medications to MSK and Radiation Therapy: Carboplatin + Pemetrexed (1,184 patients), Osimertinib (1,062), Pembrolizumab (745), Gemcitabine (708), Cisplatin + Pemetrexed (685), Carboplatin + Pembrolizumab + Pemetrexed (624), Nivolumab (533).
Notes: A correct answer must: restrict to Non-Small Cell Lung Cancer patients in msk_chord_2024, use the treatment timeline data (clinical_event_derived + clinical_event_data AGENT/SUBTYPE, see treatment-guide), and report the most frequent regimens with patient counts; Carboplatin + Pemetrexed and Osimertinib should lead. Listing single agents instead of same-day combinations is partial credit. Must not: invent regimens, report treatments for all cancer types, or count Radiation Therapy or Prior Medications to MSK rows as systemic regimens.

#19 DataAlteration frequency · msk_chord_2024What are the top 10 most frequently mutated genes across all cancer types in the MSK-CHORD Study?

Reference answer (checked 2026-09-23): TP53 52.4%, KRAS 28.5%, APC 19.1%, PIK3CA 14.8%, EGFR 8.6%, ARID1A 7.4%, SMAD4 7.2%, KMT2D 7.1%, KMT2C 6.5%, ATM 5.5% (of 25,040 profiled samples).
Expected links: https://www.cbioportal.org/study/summary?id=msk_chord_2024

#20 DataAlteration frequency · msk_chord_2024What percentage of colorectal cancer samples have KRAS mutations in the MSK-CHORD Study?

Reference answer (checked 2026-09-23): In the MSK-CHORD (MSK Nature 2024) cohort, KRAS is mutated in 2,355 of 5,543 colorectal cancer samples, corresponding to a mutation frequency of 42.5%.
Expected links: https://www.cbioportal.org/study/summary?id=msk_chord_2024#filterJson={
https://www.cbioportal.org/results/cancerTypesSummary?cancer_study_list=msk_chord_2024&case_set_id=msk_chord_2024_all&gene_list=KRAS%253A%2520MUT

#21 DataCo-occurrence & exclusivity · msk_chord_2024What are the most commonly co-occurring mutation pairs in breast cancer samples from the MSK-CHORD Study?

Reference answer (checked 2026-09-23): Most frequent co-mutated pairs in 5,368 breast cancer samples: PIK3CA+TP53 597 (11.1%), CDH1+PIK3CA 346 (6.4%), MAP3K1+PIK3CA 256 (4.8%), GATA3+PIK3CA 230 (4.3%), PIK3CA+KMT2C 218 (4.1%).
Notes: These are raw co-occurrence counts. If the answer claims statistically significant co-occurrence, it must hand off to cBioPortal's Mutual Exclusivity tab rather than invent p-values.

#22 AnalysisSurvival & outcomes · msk_chord_2024Do patients with PIK3CA mutations have different overall survival outcomes compared to PIK3CA wild-type patients in breast cancer from the MSK-CHORD Study?

Expected links: https://www.cbioportal.org/results/comparison/survival?session_id=6936f6a2b2bb32147b0176f1

#23 AnalysisCohort & clinical counts · msk_chord_2024What is the correlation between tumor mutational burden and microsatellite instability status in colorectal cancer patients from the MSK-CHORD Study?

Reference answer (checked 2026-09-23): Tumor mutational burden (TMB) demonstrates a significant positive association with microsatellite instability (MSI) in colorectal cancer patients from the MSK-CHORD cohort. Using continuous MSI scores, TMB correlates moderately with MSI, with a Spearman coefficient of 0.37 (p = 1.18 × 10⁻¹⁷⁵) and a Pearson coefficient of 0.58 (p = 0.00), indicating that higher MSI scores correspond to higher mutational burden. Consistent with these quantitative correlations, categorical MSI typing shows a clear separation wherein MSI-instable tumors exhibit high TMB, MSI-stable tumors show uniformly low TMB, and indeterminate cases fall between these groups.
Expected links: https://www.cbioportal.org/study/plots?id=msk_chord_2024&plots_horz_selection=%7B%22dataType%22%3A%22clinical_attribute%22%2C%22selectedDataSourceOption%22%3A%22TMB_NONSYNONYMOUS%22%7D&plots_vert_selection=%7B%22dataType%22%3A%22clinical_attribute%22%2C%22selectedDataSourceOption%22%3A%22MSI_SCORE%22%7D#filterJson={

#24 AnalysisTreatment · msk_chord_2024Which genomic alterations are associated with immunotherapy response in melanoma patients from the MSK-CHORD Study?

Reference answer (checked 2026-09-24): MSK-CHORD contains no melanoma patients (its five cancer types are NSCLC, colorectal, breast, prostate and pancreatic cancer) and has no immunotherapy response variable, so this cannot be answered from MSK-CHORD.
Notes: A correct answer must: state that MSK-CHORD has no melanoma patients (only NSCLC, colorectal, breast, prostate and pancreatic cancer) and no direct immunotherapy response variable, then decline or offer an alternative (e.g. a melanoma immunotherapy cohort in another study, or the same question for an MSK-CHORD cancer type). Must not: invent melanoma patients, genomic associations, response rates or p-values for MSK-CHORD.

#25 AnalysisAlteration frequency · msk_chord_2024How does the mutation landscape differ between primary and metastatic samples from the same patients in the MSK-CHORD Study?

Reference answer (checked 2026-09-24): Paired primary/metastasis data is essentially unavailable in MSK-CHORD: 24,950 patients have 25,040 samples, only 90 patients have more than one sample, and only 26 patients have both a Primary and a Metastasis sample.
Notes: No cBioPortal page shows this, so a link is not expected. A correct answer must: say that a within-patient comparison is essentially impossible (only 26 patients have both a Primary and a Metastasis sample; 90 have more than one sample). An unpaired comparison of all primary vs all metastatic samples (SAMPLE_TYPE, via group comparison) is acceptable only if clearly labelled as unpaired. Must not: present an unpaired comparison as if it were within the same patients, or imply many patients have paired samples.

#26 DataAlteration frequency · msk_chord_2024What percentage of genomic events in the MSK-CHORD Study occur in genes that are off-panel (not covered by the sequencing panel used)?

Reference answer (checked 2026-09-23): About 1.0% of genomic events (2,877 of 279,290) are off-panel: 150 of 208,232 mutations (0.07%), 0 of 60,787 CNAs, and 2,727 of 10,271 structural variants (26.6%).
Notes: The key point is that off-panel events are rare for mutations and CNAs and concentrated in structural variants.

#27 Out of scopeOut of scope · msk_chord_2024Is there a correlation between ERBB2 gene amplification and ERBB2 protein expression levels in breast cancer samples from the MSK-CHORD Study?

Notes: A correct answer must: say MSK-CHORD has no protein expression data (no RPPA/protein profile), so ERBB2 protein levels can't be correlated; may offer ERBB2 amplification vs the patient-level clinical HER2 status attribute instead. Must not: invent a correlation coefficient.

#28 DataAlteration frequency · gbm_tcga_pan_can_atlas_2018"What percentage of glioblastoma patients have alterations in RB pathway genes (CDKN2A, CDK4, RB1)?"

Reference answer (checked 2026-09-23): 80.4% of patients (304 of 378 profiled for mutations and CNA) have an RB pathway alteration: CDKN2A 57.4% (almost all deep deletions), CDK4 15.9% (amplifications), RB1 12.4% (mostly mutations).
Expected links: https://www.cbioportal.org/results/oncoprint?cancer_study_list=gbm_tcga_pan_can_atlas_2018&case_set_id=gbm_tcga_pan_can_atlas_2018_cnaseq&gene_list=CDKN2A%20CDK4%20RB1
Notes: Must include copy-number alterations (CDKN2A deletion, CDK4 amplification), not mutations only.

#29 AnalysisCo-occurrence & exclusivity · gbm_tcga_pan_can_atlas_2018"Are mutations in CDKN2A, CDK4, and RB1 mutually exclusive in glioblastoma patients?"

Reference answer (checked 2026-09-23): Point mutations in CDKN2A and CDK4 are rare in GBM; their alterations are deep deletions and amplifications. The alterations tend toward mutual exclusivity: CDK4 and RB1 alterations never co-occur; CDKN2A overlaps RB1 in 7 and CDK4 in 13 patients.
Notes: A correct answer must: point out that CDKN2A and CDK4 alterations in GBM are mostly copy-number events rather than point mutations, describe the tendency toward mutual exclusivity, and hand off significance to the Mutual Exclusivity tab. Must not: invent p-values or log odds ratios.

#30 AnalysisExpression & multi-omics · gbm_tcga_pan_can_atlas_2018In the TCGA PanCancer Atlas glioblastoma study, is CDK4 mRNA expression significantly higher in samples with CDK4 amplification compared to diploid samples?

Reference answer (checked 2026-09-23): Yes in direction: CDK4 mRNA is higher in GISTIC-amplified (2) than diploid (0) samples in gbm_tcga_pan_can_atlas_2018. Significance comes from cBioPortal Plots / group comparison or an explicitly named test.
Notes: A correct answer must: compare CDK4 mRNA between GISTIC amplified (2) and diploid (0) samples in gbm_tcga_pan_can_atlas_2018, state the direction, and hand off the significance test to cBioPortal Plots / group comparison (or compute it explicitly with a named test). Must not: invent a p-value.

#31 DataAlteration frequency · All TCGA PancanWhich cancer types have the highest frequency of EGFR mutations across all TCGA Pan-Cancer Atlas studies?

Reference answer (checked 2026-09-23): Across TCGA Pan-Cancer Atlas studies, glioblastoma has by far the highest frequency of EGFR mutations (23.7%, 94/397 profiled samples), followed by lung adenocarcinoma (12.4%), melanoma (8.0%), endometrial cancer (7.0%) and lower grade glioma (6.8%). Stomach adenocarcinoma (4.8%) and lung squamous, colorectal, esophageal, head and neck, cervical, adrenocortical and bladder cancers (~2-3%) follow.
Expected links: https://www.cbioportal.org/results/cancerTypesSummary?cancer_study_list=laml_tcga_pan_can_atlas_2018%2Cacc_tcga_pan_can_atlas_2018%2Cblca_tcga_pan_can_atlas_2018%2Clgg_tcga_pan_can_atlas_2018%2Cbrca_tcga_pan_can_atlas_2018%2Ccesc_tcga_pan_can_atlas_2018%2Cchol_tcga_pan_can_atlas_2018%2Ccoadread_tcga_pan_can_atlas_2018%2Cdlbc_tcga_pan_can_atlas_2018%2Cesca_tcga_pan_can_atlas_2018%2Cgbm_tcga_pan_can_atlas_2018%2Chnsc_tcga_pan_can_atlas_2018%2Ckich_tcga_pan_can_atlas_2018%2Ckirc_tcga_pan_can_atlas_2018%2Ckirp_tcga_pan_can_atlas_2018%2Clihc_tcga_pan_can_atlas_2018%2Cluad_tcga_pan_can_atlas_2018%2Clusc_tcga_pan_can_atlas_2018%2Cmeso_tcga_pan_can_atlas_2018%2Cov_tcga_pan_can_atlas_2018%2Cpaad_tcga_pan_can_atlas_2018%2Cpcpg_tcga_pan_can_atlas_2018%2Cprad_tcga_pan_can_atlas_2018%2Csarc_tcga_pan_can_atlas_2018%2Cskcm_tcga_pan_can_atlas_2018%2Cstad_tcga_pan_can_atlas_2018%2Ctgct_tcga_pan_can_atlas_2018%2Cthym_tcga_pan_can_atlas_2018%2Cthca_tcga_pan_can_atlas_2018%2Cucs_tcga_pan_can_atlas_2018%2Cucec_tcga_pan_can_atlas_2018%2Cuvm_tcga_pan_can_atlas_2018&Z_SCORE_THRESHOLD=2.0&RPPA_SCORE_THRESHOLD=2.0&profileFilter=mutations%2Cstructural_variants%2Cgistic&case_set_id=all&gene_list=EGFR%253AMUT

#32 DataVariants & hotspots · luad_tcga_pan_can_atlas_2018What are the most frequent EGFR mutation variants in lung adenocarcinoma and what percentage are known hotspot mutations?

Reference answer (checked 2026-09-23): EGFR is mutated in 70 of 566 samples (12.4%, 86 mutations). Most frequent: L858R (23), E746_A750del (16), L861Q (3), E709_T710delinsD (3). About 64% of EGFR mutations (55/86) are known activating hotspots (L858R, exon 19 deletions, L861Q, G719X, S768I, T790M, exon 20 insertions).
Expected links: https://www.cbioportal.org/results/mutations?cancer_study_list=luad_tcga_pan_can_atlas_2018&case_set_id=luad_tcga_pan_can_atlas_2018_sequenced&gene_list=EGFR
Notes: The hotspot share depends on the hotspot definition; L858R and exon 19 deletions as the top variants is the key point.

#33 AnalysisExpression & multi-omics · ov_tcga_pan_can_atlas_2018What is the correlation coefficient between EGFR copy number and EGFR mRNA expression in ovarian cancer?

Reference answer (checked 2026-09-23): Moderate positive correlation: Spearman ρ ≈ 0.36 (Pearson ≈ 0.34) between EGFR copy number (log2 CNA) and EGFR mRNA (RNA-seq, log2) across 295 samples.
Notes: Small differences from other expression/CN profile choices are fine; must name the correlation method.

#34 AnalysisExpression & multi-omics · ov_tcga_pan_can_atlas_2018In the TCGA PanCancer Atlas ovarian cancer study, what is the correlation between EGFR mRNA expression and EGFR protein (RPPA) levels?

Reference answer (checked 2026-09-23): In ovarian cancer samples from the Ovarian Serous Cystadenocarcinoma (TCGA, PanCancer Atlas) study, EGFR mRNA expression and EGFR protein levels show a moderate positive correlation. The Spearman correlation is 0.45 (p = 2.96 × 10⁻¹³) and the Pearson correlation is 0.51 (p = 6.94 × 10⁻¹⁷), indicating that higher EGFR mRNA levels are generally associated with higher EGFR protein abundance.
Expected links: https://www.cbioportal.org/results/plots?cancer_study_list=ov_tcga_pan_can_atlas_2018&case_set_id=ov_tcga_pan_can_atlas_2018_all&gene_list=EGFR&plots_horz_selection=%7B%22dataType%22%3A%22MRNA_EXPRESSION%22%7D&plots_vert_selection=%7B%22dataType%22%3A%22PROTEIN_LEVEL%22%7D

#35 AnalysisExpression & multi-omics · ov_tcga_pan_can_atlas_2018In the TCGA PanCancer Atlas ovarian cancer study, do samples with TP53 truncating mutations have significantly lower TP53 mRNA expression compared to wild-type samples?

Reference answer (checked 2026-09-23): TP53 mRNA is lower in samples with truncating TP53 mutations than in TP53 wild-type samples in ov_tcga_pan_can_atlas_2018. Most ovarian tumors are TP53-mutant, so the wild-type group is small.
Notes: A correct answer must: compare TP53 mRNA in samples with truncating TP53 mutations vs TP53 wild-type in ov_tcga_pan_can_atlas_2018, note the small wild-type group, state the direction, and hand off significance to Plots / group comparison. Must not: invent a p-value.

#36 AnalysisExpression & multi-omics · ov_tcga_pan_can_atlas_2018Is BRCA1 promoter methylation associated with decreased BRCA1 mRNA expression in ovarian cancer?

Reference answer (checked 2026-09-23): “In the Ovarian Serous Cystadenocarcinoma (TCGA, PanCancer Atlas) cohort, BRCA1 promoter methylation is clearly associated with reduced BRCA1 mRNA expression. Increasing methylation at the promoter-associated probe (cg04658354) shows a strong inverse correlation with BRCA1 transcript levels (Pearson r = –0.69, Spearman r = –0.43; both p < 1×10⁻¹⁴). These findings indicate that promoter hypermethylation actively corresponds to transcriptional downregulation of BRCA1 in ovarian tumors.”
Expected links: https://www.cbioportal.org/study/plots?id=ov_tcga_pan_can_atlas_2018&plots_horz_selection=%7B%22selectedGeneOption%22%3A672%2C%22dataType%22%3A%22MRNA_EXPRESSION%22%2C%22logScale%22%3A%22true%22%7D&plots_vert_selection=%7B%22selectedGenericAssayOption%22%3A%22cg04658354%22%2C%22dataType%22%3A%22METHYLATION%22%2C%22logScale%22%3A%22true%22%7D
Notes: Pearson depends on the mRNA transform (-0.42 linear, -0.74 log2); the Spearman value and the inverse correlation are what matter.

#37 AnalysisSurvival & outcomes · ov_tcga_pan_can_atlas_2018Do ovarian cancer patients with BRCA1 or BRCA2 alterations have significantly different overall survival compared to wild-type patients?

Notes: A correct answer must: define BRCA1/BRCA2-altered vs unaltered groups in ov_tcga_pan_can_atlas_2018 and hand off overall survival comparison (Kaplan-Meier, log-rank) to cBioPortal group comparison / survival tab with a link; may note literature showing better survival for BRCA-mutant ovarian cancer. Must not: invent a median survival, hazard ratio or p-value.

#38 AnalysisSurvival & outcomes · luad_tcga_pan_can_atlas_2018Do lung adenocarcinoma patients with high EGFR mRNA expression (top quartile) have different survival outcomes than those with low expression?

Notes: A correct answer must: define high (top quartile) vs low EGFR mRNA groups in luad_tcga_pan_can_atlas_2018 and hand off the survival comparison to cBioPortal group comparison with a link. Must not: invent median survival, hazard ratios or p-values.

#39 AnalysisSurvival & outcomes · luad_tcga_pan_can_atlas_2018"What are the survival differences between EGFR-mutated, EGFR-amplified, and EGFR wild-type lung adenocarcinoma patients?"

Reference answer (checked 2026-09-24): “Across patients from the Lung Adenocarcinoma (TCGA, PanCancer Atlas) study, EGFR mutations were not associated with a statistically significant difference in overall survival compared with EGFR–wild-type tumors (log-rank p = 0.125). Median overall survival was 42.5 months for EGFR-mutant patients versus 53.3 months for wild-type patients, with an estimated hazard ratio (HR) of 1.38 (95% CI: 0.87–2.18), indicating no clear survival disadvantage. In contrast, EGFR amplification was associated with significantly poorer survival (p = 2.82 × 10⁻³). EGFR-amplified tumors showed a markedly reduced median overall survival of 36.7 months compared with 53.3 months for EGFR–wild-type patients. The hazard ratio for amplified tumors was 2.15 (95% CI: 1.05–4.41), consistent with substantially shorter survival in this subgroup. Overall, EGFR-mutant lung adenocarcinomas do not exhibit significantly different survival outcomes compared with wild-type cases in this cohort, whereas EGFR-amplified tumors show significantly worse survival.”
Expected links: https://www.cbioportal.org/results/comparison/survival?cancer_study_list=luad_tcga_pan_can_atlas_2018&tab_index=tab_visualize&profileFilter=mutations%2Cgistic&case_set_id=luad_tcga_pan_can_atlas_2018_all&gene_list=EGFR%253A%2520MUT%253B%250AEGFR%253A%2520AMP%253B%250AEGFR&comparison_selectedGroups=%5B%22Unaltered%20group%22%2C%22EGFR%3A%20AMP%22%2C%22EGFR%3A%20MUT%22%5D
Notes: The reference statistics are context from cBioPortal's survival comparison, not required output (KM medians: EGFR MUT n=65 42.5 mo, EGFR AMP n=25 36.7 mo, unaltered n=430 52.6 mo; AMP vs unaltered log-rank p=0.0032). The agent cannot run Kaplan-Meier or log-rank tests. Group comparison links are session-based (`comparisonId` differs on every run): judge a comparison link by the groups and view the rendered page shows, not by matching the reference id or URL. A correct answer must: define the three groups (EGFR mutated, EGFR amplified, unaltered) in luad_tcga_pan_can_atlas_2018 and give a working cBioPortal survival comparison link for them (e.g. the expected link), handing off the statistics to cBioPortal. Must not: invent p-values, hazard ratios or medians, or report a median computed from raw OS_MONTHS values.

#40 DataAlteration frequency · luad_tcga_pan_can_atlas_2018What are the most frequently altered genes in KRAS wild-type lung adenocarcinoma patients?

Reference answer (checked 2026-09-23): In the 398 KRAS-wild-type (no KRAS mutation) samples: TP53 58.5%, TTN 47.7%, MUC16 39.7%, CSMD3 39.4%, RYR2 36.9%, LRP1B 33.2%.
Notes: Answers that exclude long passenger-prone genes (TTN, MUC16, CSMD3) and highlight TP53, EGFR, STK11, KEAP1 are acceptable if they say so.

#41 AnalysisExpression & multi-omics · lusc_tcga_pan_can_atlas_2018How does PTEN alteration (mutations or homozygous deletions) affect pAKT protein levels in lung squamous cell carcinoma?

Reference answer (checked 2026-09-23): PTEN mutations or homozygous deletions are associated with significantly increased pAKT protein levels in LUSC (e.g., AKT1_pT308: 0.83 vs 0.17; log2 ratio = 0.66; p = 2.6 × 10⁻⁸) according to the Lung Squamous Cell Carcinoma (TCGA, PanCancer Atlas) study.
Expected links: https://www.cbioportal.org/results/comparison/protein?cancer_study_list=lusc_tcga_pan_can_atlas_2018&case_set_id=lusc_tcga_pan_can_atlas_2018_all&gene_list=PTEN%253A%2520MUT%2520HOMDEL%253B

#42 DataCo-occurrence & exclusivity · ucec_tcga_pan_can_atlas_2018In the TCGA PanCancer Atlas endometrial cancer study, what percentage of patients have co-occurring oncogenic mutations in both KRAS and NRAS?

Expected links: https://www.cbioportal.org/results/mutualExclusivity?cancer_study_list=ucec_tcga_pan_can_atlas_2018&case_set_id=ucec_tcga_pan_can_atlas_2018_all&gene_list=KRAS%253A%2520DRIVER%253B%250ANRAS%253A%2520DRIVER%253B
Notes: The reference figure is context, not required output: with cBioPortal's OncoKB/hotspot driver annotations, 2 of 511 patients (0.39%) have both KRAS and NRAS driver mutations (mutual exclusivity p = 1). This database doesn't store driver annotations for this study (driver_filter is empty), so the agent can't compute it from SQL. A correct answer must: either give a cBioPortal link using the OQL DRIVER modifier for KRAS and NRAS in ucec_tcga_pan_can_atlas_2018 (e.g. the expected link) and say the driver-only co-occurrence is shown there, or report the driver-only figure (~0.4%, 2 of ~511 patients). A count over all KRAS/NRAS mutations (~4 samples) is fine as context when labeled as including non-driver mutations. Must not: present an all-mutation count as oncogenic-only, or claim a significant association.

#43 AnalysisVariants & hotspots · ucec_tcga_pan_can_atlas_2018In the TCGA PanCancer Atlas endometrial cancer study, in samples with both KRAS and NRAS mutations, what is the distribution of variant allele frequencies suggesting clonal vs subclonal events?

Reference answer (checked 2026-09-23): Only 4 samples have both KRAS and NRAS mutations: TCGA-A5-A0G2 (KRAS Q61H VAF 0.52; NRAS D47N 0.29, F78S 0.21), TCGA-AX-A0J0 (KRAS K176Q 0.42; NRAS E162* 0.10), TCGA-B5-A0JV (KRAS G12D 0.20; NRAS Q61K 0.19), TCGA-DF-A2KZ (KRAS A146V 0.22; NRAS Q61R 0.25). In the first two, the lower NRAS VAFs suggest subclonal NRAS events; in the last two the VAFs are similar.
Notes: Must not overinterpret clonality without purity/copy-number adjustment.

#44 AnalysisCohort & clinical counts · ucec_tcga_pan_can_atlas_2018What percentage of endometrial cancer samples have hypermutation (>5000 mutations) and how does this correlate with histological subtype?

Reference answer (checked 2026-09-23): According to the Uterine Corpus Endometrial Carcinoma (TCGA, PanCancer Atlas) study, approximately 7–8% of endometrial cancer samples are hypermutated (>5000 mutations). These hypermutated tumors show strong enrichment for the POLE subtype and, to a lesser extent, MSI tumors, whereas non-hypermutated tumors are primarily CN-high or CN-low.
Expected links: https://www.cbioportal.org/comparison/clinical?comparisonId=692051a8b2bb32147b014312

#45 AnalysisAlteration frequency · ucec_tcga_pan_can_atlas_2018What are the most frequently mutated genes in copy-number high subtype endometrial cancers compared to other subtypes?

Reference answer (checked 2026-09-23): In copy-number-high (serous-like) endometrial cancers (163 patients) vs other subtypes (344): TP53 87% vs 15%, PPP2R1A 31% vs 10%, PIK3CA 33% vs 58%, FBXW7 18% vs 19%, while PTEN (17% vs 88%) and ARID1A are much less frequent in CN-high.
Notes: Key points: TP53 dominates CN-high; PTEN/PIK3CA/ARID1A are enriched in the other subtypes.

#46 DataVariants & hotspots · All TCGA PancanWhich cancer types show the highest frequency of BRAF V600E mutations across all TCGA Pan-Cancer Atlas studies?

Reference answer (checked 2026-09-23): Across TCGA Pan-Cancer Atlas studies, thyroid carcinoma has by far the highest frequency of BRAF V600E (58.0% of profiled samples), followed by melanoma (35.9%) and colorectal cancer (9.0%). Cholangiocarcinoma (2.8%, 1/36), lung adenocarcinoma (1.6%) and glioblastoma (1.3%) follow; all other cancer types are below 1%.
Expected links: https://www.cbioportal.org/results/cancerTypesSummary?cancer_study_list=laml_tcga_pan_can_atlas_2018%2Cacc_tcga_pan_can_atlas_2018%2Cblca_tcga_pan_can_atlas_2018%2Clgg_tcga_pan_can_atlas_2018%2Cbrca_tcga_pan_can_atlas_2018%2Ccesc_tcga_pan_can_atlas_2018%2Cchol_tcga_pan_can_atlas_2018%2Ccoadread_tcga_pan_can_atlas_2018%2Cdlbc_tcga_pan_can_atlas_2018%2Cesca_tcga_pan_can_atlas_2018%2Cgbm_tcga_pan_can_atlas_2018%2Chnsc_tcga_pan_can_atlas_2018%2Ckich_tcga_pan_can_atlas_2018%2Ckirc_tcga_pan_can_atlas_2018%2Ckirp_tcga_pan_can_atlas_2018%2Clihc_tcga_pan_can_atlas_2018%2Cluad_tcga_pan_can_atlas_2018%2Clusc_tcga_pan_can_atlas_2018%2Cmeso_tcga_pan_can_atlas_2018%2Cov_tcga_pan_can_atlas_2018%2Cpaad_tcga_pan_can_atlas_2018%2Cpcpg_tcga_pan_can_atlas_2018%2Cprad_tcga_pan_can_atlas_2018%2Csarc_tcga_pan_can_atlas_2018%2Cskcm_tcga_pan_can_atlas_2018%2Cstad_tcga_pan_can_atlas_2018%2Ctgct_tcga_pan_can_atlas_2018%2Cthym_tcga_pan_can_atlas_2018%2Cthca_tcga_pan_can_atlas_2018%2Cucs_tcga_pan_can_atlas_2018%2Cucec_tcga_pan_can_atlas_2018%2Cuvm_tcga_pan_can_atlas_2018&profileFilter=mutations%2Cstructural_variants%2Cgistic&case_set_id=all&gene_list=BRAF%253A%2520MUT%253DV600E

#47 AnalysisExpression & multi-omics · All TCGA PancanHow does ERBB2 mRNA expression vary across different cancer types in TCGA Pan-Cancer Atlas studies?

Notes: A correct answer must: use the TCGA PanCancer Atlas studies (pan_cancer_tcga), compare ERBB2 mRNA by cancer type (e.g. via cBioPortal Plots or a per-type summary), and identify breast cancer as highest overall (driven by HER2-amplified tumors) with upper GI (stomach, esophageal) and bladder also high. Must not: combine non-TCGA studies or invent values.

#48 AnalysisExpression & multi-omics · brca_tcga_pan_can_atlas_2018In the TCGA PanCancer Atlas breast cancer study, what is the concordance between ERBB2 copy number amplification, mRNA overexpression, and protein overexpression?

Reference answer (checked 2026-09-23): In brca_tcga_pan_can_atlas_2018, ERBB2 amplification (GISTIC 2) largely coincides with mRNA overexpression (z-score > 2) and protein overexpression (RPPA z-score) among samples with all three data types.
Notes: A correct answer must: define ERBB2 amplification (GISTIC 2), mRNA overexpression (z-score threshold, e.g. > 2) and protein overexpression (RPPA z-score) in brca_tcga_pan_can_atlas_2018, report the overlap/concordance among samples with all three data types, and reach the reference conclusion. Must not: invent percentages.

#49 AnalysisExpression & multi-omics · All TCGA PancanWhich cancer types have the highest aneuploidy scores and how does this correlate with mutation burden across TCGA Pan-Cancer studies?

Reference answer (checked 2026-09-23): Across TCGA Pan-Cancer studies, tumors with the highest aneuploidy burdens included seminoma, non-seminomatous germ cell tumor, adrenocortical carcinoma, non–small cell lung cancer, and bladder cancer, all of which displayed elevated median aneuploidy scores. Although bladder cancer and non–small cell lung cancer also ranked among the tumor types with the highest mutational burdens, seminoma and non-seminomatous germ cell tumors showed some of the lowest TMB values, with adrenocortical carcinoma exhibiting intermediate levels. These patterns indicate that aneuploidy does not reliably coincide with high mutation loads across tumor types.
Expected links: https://www.cbioportal.org/study/plots?id=laml_tcga_pan_can_atlas_2018%2Cacc_tcga_pan_can_atlas_2018%2Cblca_tcga_pan_can_atlas_2018%2Clgg_tcga_pan_can_atlas_2018%2Cbrca_tcga_pan_can_atlas_2018%2Ccesc_tcga_pan_can_atlas_2018%2Cchol_tcga_pan_can_atlas_2018%2Ccoadread_tcga_pan_can_atlas_2018%2Cdlbc_tcga_pan_can_atlas_2018%2Cesca_tcga_pan_can_atlas_2018%2Cgbm_tcga_pan_can_atlas_2018%2Chnsc_tcga_pan_can_atlas_2018%2Ckich_tcga_pan_can_atlas_2018%2Ckirc_tcga_pan_can_atlas_2018%2Ckirp_tcga_pan_can_atlas_2018%2Clihc_tcga_pan_can_atlas_2018%2Cluad_tcga_pan_can_atlas_2018%2Clusc_tcga_pan_can_atlas_2018%2Cmeso_tcga_pan_can_atlas_2018%2Cov_tcga_pan_can_atlas_2018%2Cpaad_tcga_pan_can_atlas_2018%2Cpcpg_tcga_pan_can_atlas_2018%2Cprad_tcga_pan_can_atlas_2018%2Csarc_tcga_pan_can_atlas_2018%2Cskcm_tcga_pan_can_atlas_2018%2Cstad_tcga_pan_can_atlas_2018%2Ctgct_tcga_pan_can_atlas_2018%2Cthym_tcga_pan_can_atlas_2018%2Cthca_tcga_pan_can_atlas_2018%2Cucs_tcga_pan_can_atlas_2018%2Cucec_tcga_pan_can_atlas_2018%2Cuvm_tcga_pan_can_atlas_2018&plots_horz_selection=%7B%22dataType%22%3A%22clinical_attribute%22%2C%22selectedDataSourceOption%22%3A%22CANCER_TYPE%22%2C%22logScale%22%3A%22true%22%7D&plots_vert_selection=%7B%22dataType%22%3A%22clinical_attribute%22%2C%22selectedDataSourceOption%22%3A%22ANEUPLOIDY_SCORE%22%2C%22logScale%22%3A%22false%22%7D

#50 AnalysisAlteration frequency · All TCGA Pancan"Are mutations in DNA repair pathway genes (BRCA1, BRCA2, ATM, CHEK2) enriched in specific cancer types across TCGA Pan-Cancer Atlas?"

Notes: A correct answer must: compute per-cancer-type mutation frequency of BRCA1/BRCA2/ATM/CHEK2 in pan_cancer_tcga (e.g. gene_mutation_frequency_by_cancer_type) and note that hypermutated types (endometrial, colorectal MSI, melanoma) inflate frequencies; may hand off enrichment tests to group comparison. Must not: sum across studies or invent p-values.

#51 DataCohort & clinical counts · nbl_target_2018_pubWhat fraction of patients were older than five when diagnosed according to the Pediatric Neuroblastoma study from TARGET?

Reference answer (checked 2026-09-23): 11.62%
Expected links: https://www.cbioportal.org/study/summary?id=nbl_target_2018_pub

#52 DataVariants & hotspots · brca_tcga_pan_can_atlas_2018What is the most frequent mutation in the TP53 gene in the TCGA breast cancer study?

Reference answer (checked 2026-09-23): R175H, found in 21 of 1,066 profiled samples (2.0%), i.e. 6.1% of TP53-mutated samples and 5.95% of all TP53 mutations
Expected links: https://www.cbioportal.org/results/mutations?cancer_study_list=brca_tcga_pan_can_atlas_2018&case_set_id=brca_tcga_pan_can_atlas_2018_all&gene_list=TP53

#53 DataAlteration frequency · luad_tcga_pan_can_atlas_2018How many patients have an EGFR amplification in the TCGA Lung Adenocarcinoma study?

Reference answer (checked 2026-09-23): 26 patients (5.1% of CNA-profiled samples) have an EGFR amplification
Expected links: https://www.cbioportal.org/study/summary?id=luad_tcga_pan_can_atlas_2018

#54 DataVariants & hotspots · coadread_tcga_pan_can_atlas_2018Which KRAS mutations are most common in colorectal cancer?

Reference answer (checked 2026-09-23): G12D (58 samples); G12V (49); G13D (37); A146T (16); G12C (15); G12A (10); G12S (8)
Expected links: https://www.cbioportal.org/results/mutations?cancer_study_list=coadread_tcga_pan_can_atlas_2018&case_set_id=coadread_tcga_pan_can_atlas_2018_all&gene_list=KRAS

#55 DataSurvival & outcomes · nbl_target_2018_pubWhat is the median survival time in the Pediatric Neuroblastoma study from TARGET?

Reference answer (checked 2026-09-24): Median overall survival is not reached: the Kaplan-Meier curve for nbl_target_2018_pub (1,072 patients with OS data, 397 deaths) levels off at about 55% survival and never falls below 50%, so the median cannot be estimated.
Expected links: https://www.cbioportal.org/study/summary?id=nbl_target_2018_pub
Notes: A correct answer must: use nbl_target_2018_pub and say the Kaplan-Meier median is not reached (or hand off to the study's KM plot), linking the study summary page. Must not: report the raw median of OS_MONTHS (or an average) as the median survival, or use a different TARGET study.

#56 AnalysisTreatment · gbm_tcga_pub2013In the TCGA glioblastoma study (Cell 2013), how does methylation of the MGMT gene promoter affect the prognosis and treatment response in patients with glioblastoma?

Notes: The reference statistics are context from cBioPortal's survival comparison, not required output: MGMT promoter methylation is associated with improved overall survival across multiple treatment contexts in glioblastoma according to the Glioblastoma (TCGA, Cell 2013) study. Patients with methylated tumors receiving standard radiation show a median survival of 21.2 months (95% CI 7.0–NA), and those treated with standard radiation plus temozolomide (TMZ) chemo show the longest survival at 23.5 months (22.2–47.9). Methylated tumors treated with TMZ chemoradiation plus TMZ achieve a median survival of 17.8 months (15.7–21.3), whereas those receiving unspecified radiation show markedly poorer outcomes (2.2 months, 1.4–3.9). In contrast, unmethylated tumors show consistently shorter survival across comparable regimens: 11.6 months (10.3–NA) with standard radiation, 12.7 months (11.7–19.3) with standard radiation plus TMZ, 14.5 months (12.9–15.9) with TMZ chemoradiation plus TMZ, and 2.7 months (1.2–5.4) with unspecified radiation. Across all groups, survival curves for methylated tumors cluster at higher survival times, indicating that MGMT promoter methylation confers both a prognostic advantage and increased benefit from TMZ-containing therapy. The agent cannot run Kaplan-Meier or log-rank tests. Group comparison links are session-based (`comparisonId` differs on every run): judge a comparison link by the groups and view the rendered page shows, not by matching the reference id or URL. A correct answer must: compare survival of MGMT-methylated vs unmethylated patients in gbm_tcga_pub2013 (optionally by treatment) and give a working cBioPortal survival comparison link for them, handing off the statistics to cBioPortal. Must not: invent p-values, hazard ratios or medians, or report a median computed from raw OS_MONTHS values.

#57 Out of scopeOut of scope · NAFor patients with the EML4-ALK fusion gene in lung cancer how do the different fusion variants affect their long-term quality of life and risk of developing a secondary cancer?

Reference answer (checked 2026-09-23): This question cannot be answered with the cBioPortal databse
Notes: Out of scope. A correct answer must say cBioPortal can't answer this; must not invent an answer.

#58 AnalysisSurvival & outcomes · pancan_pcawg_2020In the “Pan-cancer analysis of whole genomes” study what is the survival difference and the corresponding statistical significance between patients with mutations in both TP53 and KRAS versus patients with only a KRAS mutation?

Reference answer (checked 2026-09-23): This comparison is not meaningful in the Pan-cancer analysis of whole genomes study: only 282 patients have overall survival data, and just 1 of them has both TP53 and KRAS mutations and 3 have only a KRAS mutation, so no reliable median survival or p-value can be computed.
Expected links: https://www.cbioportal.org/comparison/survival?comparisonId=691f3fd9b2bb32147b013ee9
Notes: A correct answer must say there are too few patients with survival data and both mutations to compare; must not invent a median or p-value.

#59 Out of scopeOut of scope · NAAre there studies that were not processed using polyA enrichment in order to explore lncRNA-related questions?

Reference answer (checked 2026-09-23): Library preparation protocols are not contained in the database and should be consulted in the publications underlying the studies.
Notes: Out of scope. A correct answer must say cBioPortal can't answer this; must not invent an answer.

#60 AnalysisExpression & multi-omics · brca_tcga_pan_can_atlas_2018In the Breast Invasive Carcinoma TCGA study what are the top 5 down-regulated genes in TP53 mutated samples compared to non-mutated ones?

Reference answer (checked 2026-09-23): AGR3; TFF1; SRARP; CYP2B7P; CPB1
Expected links: https://www.cbioportal.org/results/comparison/mrna?cancer_study_list=brca_tcga_pan_can_atlas_2018&case_set_id=brca_tcga_pan_can_atlas_2018_all&gene_list=TP53

#61 AnalysisSurvival & outcomes · nbl_target_2018_pubIn the Pediatric Neuroblastoma study from TARGET what is the survival difference and the corresponding statistical significance between patients who were older than four when diagnosed and the younger ones?

Reference answer (checked 2026-09-24): Older patients (>4) had significantly worse overall survival compared to younger patients (≤4) with a higher hazard of death (HR = 1.47; 95% CI 1.17–1.86; log-rank p = 3.37×10⁻⁴). Their median overall survival was 79.0 months (95% CI 53–122) whereas median survival was not reached in the ≤4 group.
Notes: The reference statistics are context from cBioPortal's survival comparison, not required output (log-rank p=3.37e-4; KM median 79.0 months for >4 years, not reached for <=4 years). The agent cannot run Kaplan-Meier or log-rank tests. Group comparison links are session-based (`comparisonId` differs on every run): judge a comparison link by the groups and view the rendered page shows, not by matching the reference id or URL. A correct answer must: define the two groups by age at diagnosis (>4 vs <=4 years) in nbl_target_2018_pub and give a working cBioPortal survival comparison link for them, handing off the statistics to cBioPortal. Must not: invent p-values, hazard ratios or medians, or report a median computed from raw OS_MONTHS values.

#62 AnalysisCohort & clinical counts · gbm_tcga_pan_can_atlas_2018In the TCGA Glioblastoma multiforme study compare the median patient age at diagnosis between patients with IDH1 R132H mutation and patients with wild-type IDH1.

Reference answer (checked 2026-09-23): Patients with IDH1 R132H mutations are diagnosed at a substantially younger age (median ~40 years) compared with IDH1 wild-type patients (median ~60 years; Wilcoxon p = 3.9 × 10⁻⁸) according to the Glioblastoma Multiforme (TCGA, PanCancer Atlas) study.
Expected links: https://www.cbioportal.org/results/comparison/clinical?cancer_study_list=gbm_tcga_pan_can_atlas_2018&profileFilter=mutations&case_set_id=gbm_tcga_pan_can_atlas_2018_all&gene_list=IDH1%253A%2520MUT%253DR132H%253B%250AIDH1&comparison_selectedGroups=%5B%22Unaltered%20group%22%2C%22IDH1%3A%20MUT%3DR132H%22%5D
Notes: Expected link view: Select plot for “Diagnosis age”

#63 AnalysisExpression & multi-omics · brca_tcga_pan_can_atlas_2018In the TCGA PanCancer Atlas breast cancer study, what is the correlation coefficient between EGFR expression levels and PIK3CA mutation status considering only patients who also have a homozygous deletion of PTEN?

Expected links: https://www.cbioportal.org/study/plots?id=brca_tcga_pan_can_atlas_2018&plots_horz_selection=%7B%22selectedGeneOption%22%3A1956%2C%22dataType%22%3A%22MRNA_EXPRESSION%22%7D&plots_vert_selection=%7B%22selectedGeneOption%22%3A5290%2C%22dataType%22%3A%22MUTATION_EXTENDED%22%2C%22mutationCountBy%22%3A%22MutationType%22%7D&plots_coloring_selection=%7B%7D#filterJson={%22geneFilters%22:[{%22molecularProfileIds%22:[%22brca_tcga_pan_can_atlas_2018_gistic%22

#64 AnalysisVariants & hotspots · All TCGA PancanIn TCGA PanCancer Atlas, which recurrent hotspot mutations occur almost exclusively in one cancer type?

Reference answer (checked 2026-09-28): Among missense variants with at least 10 carriers across the TCGA PanCancer Atlas studies (by CANCER_TYPE), known driver hotspots that are ≥90% one cancer type include: GTF2I L424H in thymic epithelial tumors (59/61), IDH1 R132H in glioma (358/388, 92%), BRAF V600K in melanoma (35/35), EGFR L858R in non-small cell lung cancer (23/23), GNA11 Q209L and GNAQ Q209L/P in ocular melanoma, PTEN R130G, FGFR2 S252W and PPP2R1A P179R in endometrial cancer, and DNMT3A R882H, IDH2 R140Q and FLT3 D835Y in leukemia. Many other 100%-specific variants are UV-signature passengers in melanoma (e.g. olfactory receptor genes).
Notes: No cBioPortal page shows this, so a link is not expected. A correct answer must: use the TCGA PanCancer Atlas studies (pan_cancer_tcga) grouped by CANCER_TYPE, state its specificity rule (e.g. a minimum carrier count and share of carriers in one cancer type), and name several of the driver hotspots in the reference with counts from the data. Must not: present melanoma UV passenger variants as cancer-type-specific drivers without saying so, invent counts, or cite hotspots from cancer types that aren't in TCGA (e.g. KIT/PDGFRA in GIST) as TCGA results.

#65 AnalysisAlteration frequency · brca_tcga_pan_can_atlas_2018In the TCGA PanCancer Atlas breast cancer study, what is the average tumor mutational burden of patients who have at least two of the five most frequently mutated genes and how does this compare to the average TMB of all patients?

Reference answer (checked 2026-09-23): Top 5 mutated genes: TP53, PIK3CA, TTN, CDH1, GATA3. 296 patients have ≥2 of them; their mean TMB is 5.5 mut/Mb (median 2.0) vs 2.7 (median 1.3) for all 1,066 patients (TMB_NONSYNONYMOUS).
Notes: Means are skewed by hypermutated tumors (TTN is a long gene); mentioning medians or this caveat is good.

#66 AnalysisSurvival & outcomes · coadread_tcga_pan_can_atlas_2018, crc_apc_impact_2020In colorectal cancer do patients with microsatellite instability in the TCGA dataset have a similar prognosis as those in the MSK Gastroenterology 2020 study?

Notes: A correct answer must: identify MSI-high patients in coadread_tcga_pan_can_atlas_2018 and crc_apc_impact_2020 using each study's MSI attribute, note the studies' different designs and follow-up, and hand off survival comparison to cBioPortal group comparison (one per study, or across studies with caveats). Must not: invent survival statistics or claim a direct comparison is straightforward.

#67 AnalysisCo-occurrence & exclusivity · brca_tcga_pan_can_atlas_2018In the TCGA Breast Cancer study if a patient has both a PIK3CA mutation and a PTEN deletion what is their probability of having a luminal A subtype? Is this probability different from a patient with only a PIK3CA mutation?

Expected links: https://www.cbioportal.org/results/comparison/clinical?cancer_study_list=brca_tcga_pan_can_atlas_2018&profileFilter=mutations%2Cgistic&case_set_id=brca_tcga_pan_can_atlas_2018_all&gene_list=PIK3CA%253A%2520MUT%253B%250APTEN%253A%2520HOMDEL%2520HETLOSS%253B&comparison_selectedGroups=%5B%22PIK3CA%3A%20MUT%20and%20PTEN%3A%20HOMDEL%20HETLOSS%22%2C%22PIK3CA%22%5D&comparison_createdGroupsSessionId=6936fd577783fa02c7f3ac9d
Notes: Expected link view: Select plot for "Subtype"

#68 AnalysisAlteration frequency · lusc_tcga_pan_can_atlas_2018, luad_tcga_pan_can_atlas_2018What are the key genomic differences between lung adenocarcinomas and squamous cell carcinomas identified in the Pan-Lung Cancer TCGA study?

Reference answer (checked 2026-09-23): Comparing luad_tcga_pan_can_atlas_2018 and lusc_tcga_pan_can_atlas_2018: KRAS, EGFR and STK11 mutations are enriched in adenocarcinoma; squamous has more TP53 mutation, NFE2L2/KEAP1 pathway and PIK3CA alterations, CDKN2A loss, and SOX2/TP63 (3q) amplification.
Notes: A correct answer must: compare luad_tcga_pan_can_atlas_2018 and lusc_tcga_pan_can_atlas_2018 and mention several of the key differences in the reference; may hand off to group comparison with a link. Must not: invent frequencies.

#69 DataPatient & sample lookup · All StudiesWhich patients have a TP53 G199V mutation? Which are somatic vs germline?

Reference answer (checked 2026-09-23): TP53 G199V appears in 66 distinct patient IDs (180 patient entries across 67 studies; many MSK patients appear in several overlapping studies). Calls are somatic (156 records) or have no status (38); only one is labeled germline: TCGA-A2-A0SX in brca_tcga_pub.
Notes: A correct answer must deduplicate or warn about patients shared across overlapping studies and must identify the single germline call; listing all patients is not required.

#70 DataVariants & hotspots · crc_msk_2017In MSS colorectal cancer, what is the frequency of BRAF oncogenic mutations (as defined by OncoKB)? Use the 2017 MSK study for this. Can you give me a table of all these mutations, with frequency, count and denominator?

Reference answer (checked 2026-09-24): In crc_msk_2017 (MSI_STATUS; msk_impact_2017 has no MSI attribute) there are 701 MSS samples, all profiled for mutations. BRAF V600E is in 46/701 (6.6%); any BRAF mutation in 78/701 (11.1%). Other recurrent variants: D594G 7, D594N 3, T599delinsIP 2, N581S 2, G469E 2, G469A 2. The DB has no OncoKB annotations, so the oncogenic subset must be taken from cBioPortal (OQL BRAF: MUT_DRIVER) or OncoKB.
Expected links: https://www.cbioportal.org/results/mutations?cancer_study_list=crc_msk_2017&gene_list=BRAF%3A%20MUT_DRIVER
https://www.oncokb.org/gene/BRAF
Notes: A correct answer must: use crc_msk_2017 (the 2017 MSK colorectal study with MSI_STATUS), restrict to MSS samples, and give a per-variant BRAF table with count, frequency and the denominator (701 MSS samples); state that OncoKB oncogenicity cannot be filtered from the DB, and link a cBioPortal MUT_DRIVER / OncoKB view for the oncogenic subset. (Question joins the first turn with the follow-up that received the feedback.) Must not: claim OncoKB driver annotations it did not actually retrieve, or use msk_impact_2017 with an MSI filter it does not have.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/17

#71 Out of scopeOut of scope · All StudiesWhat is the current and future support for storing and analyzing germline variants in cBioPortal, compared to other alternatives?

Notes: A correct answer must: explain that cBioPortal supports germline variants (e.g. mutation_status GERMLINE, germline-only studies) and that most features (clinical correlation, study view, plots, group comparison) work the same as for somatic data; point to docs on building a germline study. Must not: describe cBioPortal as somatic-only.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/20

#72 DataStudy discovery · All StudiesIs there any study with a polygenic risk score?

Reference answer (checked 2026-09-23): No. cBioPortal does not store polygenic risk scores; no study in the public database has a polygenic risk score attribute.
Notes: A correct answer must: answer quickly and directly that no study in cBioPortal has polygenic risk scores. Must not: run a long exhaustive search before concluding, or present an unrelated clinical risk score (e.g. RISK_SCORE in sft_sysucc_2023) as a polygenic risk score.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/21

#73 DataAlteration frequency · All StudiesWhat is the frequency of TP53 mutations in lung cancer?

Notes: A correct answer must: report a percentage (altered / mutation-profiled samples), starting from the TCGA PanCancer Atlas lung studies (LUAD, LUSC); if several studies are used, give a range per study and warn that samples can overlap between cohorts. Must not: report raw sample counts only, average across studies, or give a frequency above 100%.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/23

#74 DataAlteration frequency · All StudiesWhat are the most mutated genes in lung cancer?

Notes: A correct answer must: give frequencies as altered / mutation-profiled samples using COUNT(DISTINCT sample), name the cohort used, and provide a specific cBioPortal link (study view or results view with the genes) for further exploration. Must not: report any frequency above 100%, give a bare https://www.cbioportal.org link, or narrate a made-up tool failure ("technical issue with the navigation API"). (Also covers cbioportal-navigator#53.)
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/41

#75 DataPatient & sample lookup · pog570_bcgsc_2020How many patients have a shallow deletion for SMARCA4 in the POG study?

Reference answer (checked 2026-09-23): 111 patients (111 of 570 CNA-profiled samples, ~19.5%) have a shallow deletion (HETLOSS) of SMARCA4 in pog570_bcgsc_2020; a further 4 have a deep deletion.
Notes: A correct answer must: count HETLOSS (GISTIC -1) calls for SMARCA4 in pog570_bcgsc_2020 against CNA-profiled samples. (Rephrased from the issue's description of the request.) Must not: report 0 patients or count only deep deletions (HOMDEL).
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/41

#76 Out of scopeOut of scope · All StudiesAre TP53 mutations clinically actionable?

Notes: A correct answer must: ground any actionability statement in OncoKB (cite the evidence level) or redirect to OncoKB, and state that cBioPortal cohort data alone does not establish clinical actionability. Must not: give therapeutic or clinical guidance from general LLM knowledge.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/24

#77 NavigationPatient & sample lookup · brca_hta9_htan_2022can you show me the minerva viewer for the ohsu htan sample

Notes: A correct answer must: find the OHSU HTAN study and look up its resource links (resource_sample / resource_study / resource_definition) and return the Minerva viewer URL(s). Must not: declare the request out of scope or send the user to the HTAN portal without checking cBioPortal's resource tables.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/25

#78 Out of scopeOut of scope · All Studieswhat's the best way to put clinical data into cbioportal?

Notes: A correct answer must: describe the options: submitting to the public portal (curation team, cBioPortal/datahub, file formats), running a private instance, and visualizing own data without submission (OncoPrinter, MutationMapper); ask which use case applies. Must not: say it cannot help.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/26

#79 NavigationSurvival & outcomes · All StudiesCan you please generate a survival curve for colorectal cancer patients based on the expression levels of IMPDH2?

Notes: A correct answer must: be explicit that it cannot render plots or export files, and instead give a cBioPortal link/steps where the Kaplan-Meier curve can be produced (e.g. expression-based groups in group comparison) or return summary data in text. Must not: offer to plot Kaplan-Meier curves or export a CSV file it cannot deliver.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/28

#80 DataPatient & sample lookup · All StudiesIs there a cohort of NSCLC patient samples that have Kras mutations, wild-type p53, and high expression levels of c-Myc?

Notes: A correct answer must: attempt the query (KRAS-mutant, TP53 wild-type among mutation-profiled samples, MYC high by a stated expression z-score threshold) in a named NSCLC study with mutation + mRNA data; if a step fails, explain which part failed and suggest a workaround (e.g. querying the criteria separately). Must not: return an unexplained error.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/35

#81 DataAlteration frequency · lgggbm_tcga_pubtell me about IDH1 mutations in glioma

Notes: A correct answer must: use a pan-glioma cohort (e.g. lgggbm_tcga_pub) or clearly label a narrower cohort; use current OncoTree codes (GB, not the deprecated GBM) and note that glioblastoma is IDH-wildtype by definition. (Also covers cbioportal-mcp#36.) Must not: present a GBM-only study (e.g. gbm_tcga_pub2013) as pan-glioma, or state "IDH1 is more common in LGG than GBM" as a finding without the definitional caveat.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/9

#82 DataTreatment · difg_glass_2019How many GLASS patients developed hypermutation after TMZ treatment?

Reference answer (checked 2026-09-24): 30 of the 35 hypermutated patients in difg_glass_2019 (HYPERMUTATION_STATUS = Yes) have TMZ_TREATMENT = Yes (31 with ALKYLATING_AGENT = Yes). The flags do not confirm that hypermutation arose after TMZ. difg_glass has no hypermutation attribute.
Notes: A correct answer must: name the GLASS study (difg_glass_2019, since difg_glass lacks HYPERMUTATION_STATUS) and the denominator used (35 hypermutated patients), and note that the timing (hypermutation after TMZ) cannot be confirmed from the flags. If it lists individual patients, they must be clickable patient view links. (Also covers cbioportal-mcp#36.) Must not: list patient IDs without links, or use the deprecated GBM code.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/31

#83 AnalysisExpression & multi-omics · All TCGA PancanHelp me rank TCGA cancer studies based on median CD3 expression

Notes: A correct answer must: flag that CD3 is ambiguous (CD3D, CD3E, CD3G) and ask, or default to CD3E and say so; state which expression metric is used and compute a true median. Must not: silently average expression across all CD3 family genes.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/37

#84 Out of scopeOut of scope · All StudiesCan you analyze thyroid hormone gene expression by diagnosis in the Pediatric Brain Tumor Atlas (PBTA, Provisional) cohort?

Reference answer: PBTA is not in the public cBioPortal; it is available on the pediatric cBioPortal (https://pedcbioportal.kidsfirstdrc.org).
Notes: A correct answer must: say PBTA is not in the public cBioPortal and redirect to the pediatric cBioPortal; if it offers a substitute study (e.g. brain_cptac_2020) it must label it as a substitute with its sample count. (Rephrased from the issue's description of the request.) Must not: silently analyze a different study, or blame access problems.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/42

#85 NavigationAlteration frequency · msk_chord_2024generate a oncoprint of kras, stk11, keap1, tp53 alterations stratified by smoking status in lung adenocarcinoma from the mskcc clincogenomic sequencing cohort. Please only select oncogenic somatic alterations and remove samples with unknown smoking status

Notes: A correct answer must: use the MSK clinicogenomic cohort (msk_chord_2024) and return an OncoPrint link with oncogenic/driver-only OQL, smoking status as a clinical track, and unknown smoking status excluded. Must not: pick a different published cohort (e.g. lung_msk_2017), or return a table instead of a link.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/51

#86 DataAlteration frequency · All Studieswhat are the most common events in her2- breast cancer?

Notes: A correct answer must: select HER2-negative samples using an actual HER2 status attribute (HER2 IHC/FISH/HER2_STATUS) and report frequencies with profiled denominators that match cBioPortal study view (no UNCALLED or off-panel mutations). (Also covers cbioportal-mcp#53.) Must not: use Luminal B (or any intrinsic subtype) as a proxy for HER2-negative.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/52

#87 DataPatient & sample lookup · All StudiesShow me cases where PALB2 or ATM have germline mutations

Notes: A correct answer must: match germline status case-insensitively (the data contains both GERMLINE and Germline). Must not: miss cases because of a case-sensitive filter.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/56

#88 Out of scopeOut of scope · All Studieswhat do idh1 mutations do?

Notes: A correct answer must: clearly separate what comes from cBioPortal data from general background knowledge, and offer cBioPortal-grounded lookups (frequencies, co-mutations, outcomes). Must not: present general biology as if it came from cBioPortal, without a disclaimer.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/57

#89 DataStudy discovery · All Studiesis there any imaging data?

Reference answer: Yes, indirectly: cBioPortal does not host images, but some studies (e.g. HTAN studies such as brca_hta9_htan_2022 and crc_hta11_htan_2021) link to external imaging viewers (Minerva) via resource links.
Notes: A correct answer must: check the resource_* tables and answer that some studies (e.g. HTAN studies) link to external imaging viewers such as Minerva, listing them. Must not: say cBioPortal has no imaging data at all.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/58

#90 DataAlteration frequency · All Studiesshow me the distribution of mutations in the tert promoter across cancer types

Notes: A correct answer must: restrict to TERT promoter (non-coding) mutations such as C228T/C250T, only in studies that profile the promoter; state whether frequencies are patient- or sample-level and prefer patient-level for prevalence. (Also covers cbioportal-mcp#70's TERT report.) Must not: count all TERT alterations, or double-count samples/patients across studies.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/65

#91 NavigationVariants & hotspots · All Studiesshow me a histogram of C228T mutations in the tert promoter across cancer types

Notes: A correct answer must: treat C228T as a promoter (non-coding) variant and either build a valid view for it or explain the limitation. Must not: generate OQL that treats C228T as a protein-coding change.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/65

#92 DataVariants & hotspots · All Studieswhat is the most prevalent TP53 mutation in uterine cancer that is not a point mutation

Notes: A correct answer must: use the correct definition of point mutation (any single-nucleotide variant: missense, nonsense, silent, splice-site SNVs), so the answer covers indels/frameshifts and other non-SNV events. Must not: equate "point mutation" with missense.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/66

#93 DataPatient & sample lookup · All StudiesFind patients IDs and samples in colorectal cancer that harbor the V600V alteration in BRAF

Notes: A correct answer must: recognize V600V as a synonymous variant, explain that synonymous variants are filtered out of most cBioPortal studies so zero hits is expected, and if it suspects a typo, ask explicitly whether V600E was meant. Must not: silently substitute V600E, or report zero without explaining why.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/66

#94 DataExpression & multi-omics · All StudiesThere is a heavily discussed driver alteration in MAP2K1 at codon 105 that significantly alters mRNA stability. Please list the expression values for the tumors that have a nucleotide change at this position

Reference answer: The premise is false: there is no known MAP2K1 codon-105 driver with a documented effect on mRNA stability, and cBioPortal does not store mRNA stability.
Notes: A correct answer must: challenge the premise early and ask for the source. (Verbatim question is truncated in the issue.) Must not: run elaborate queries and stitch together a narrative around the false premise.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/67

#95 NavigationStudy discovery · msk_chord_2024download MSK-CHORD study on Non-Small Cell Lung Cancer dataset

Notes: A correct answer must: point to how to download MSK-CHORD (study page / datahub download) and how to restrict to NSCLC. Must not: run unrelated queries (e.g. on TP53) without mentioning them.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/67

#96 AnalysisSurvival & outcomes · All StudiesIs KRAS G12C more aggressive than G12D?

Notes: A correct answer must: clarify the ambiguous term "aggressive" and hand off survival comparisons to Kaplan-Meier (cBioPortal group comparison link, or R/Python) with the summary data. Must not: report a mean as median overall survival, compute median OS without Kaplan-Meier, or invent p-values.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/68

#97 DataCo-occurrence & exclusivity · All Studiesgive me a contingency table with the number of lung cancer patients with EGFR and/or KRAS alterations

Notes: A correct answer must: return a 2x2 table with correct counts for a named cohort, and for significance point to cBioPortal's mutual exclusivity tab or an external test. Must not: claim mutual exclusivity or give a p-value without an actual test, or fabricate statistics.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/68

#98 Out of scopeOut of scope · msk_chord_2024Can you think of any flaws in the methodology used in the MSK-CHORD paper (Jee et al., Nature 2024)?

Notes: A correct answer must: decline methodology critique as out of scope, offer to explore the cBioPortal data the paper is based on (msk_chord_2024), and hold the decline if pushed. (Rephrased: the user attached the paper PDF.) Must not: critique the paper's methodology or produce presentation content.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/69

#99 Out of scopeOut of scope · msk_chord_2024Can you help me write a production-ready Python script using Bokeh to build an interactive clinicogenomic dashboard to analyzing the MSK-CHORD dataset? I have some specific requirements I can give you

Notes: A correct answer must: decline building external application code, and point to the cBioPortal API / data downloads and built-in views. Must not: generate complex external pipeline or dashboard code (Bokeh, PySpark, lifelines).
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/69

#100 DataPatient & sample lookup · All StudiesHow many samples are there that have any of these mutations in SEPHS1: p.Arg371Trp, p.Arg371Gln, p.Arg371Gly?

Notes: A correct answer must: count samples with these variants, state whether counts are samples or patients, and flag that the same biological sample/patient can appear in several studies (e.g. MSK-IMPACT, MSK-CHORD, GENIE). Must not: present cross-study counts as biologically unique samples.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/70

#101 DataAlteration frequency · nbl_msk_2023List the top 20 mutated genes in study nbl_msk_2023.

Expected links: https://www.cbioportal.org/study/summary?id=nbl_msk_2023
Notes: A correct answer must: recognize that nbl_msk_2023 exists and return its top 20 mutated genes with frequencies. (Rephrased from the user's follow-up.) Must not: claim the study does not exist or substitute another neuroblastoma study.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/80

#102 DataExpression & multi-omics · All Studiescalculate median expression of ceacam5, itgb6, muc2, tpgb and muc1 mRNA in pancreatic cancer

Notes: A correct answer must: return a non-empty answer using a pancreatic study with mRNA data (e.g. paad_tcga_pan_can_atlas_2018), state the expression metric, and flag that "tpgb" is not a HUGO symbol (likely TPBG). Must not: return an empty response.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/93

#103 DataAlteration frequency · All Studiesin salivary cancer (adenoid cystic carcinoma), what are the expected drivers ? Classify them by actionability. What about BCOR mutations, either somatic or germline ?

Reference answer (checked 2026-09-24): acc_2019 is the Adenoid Cystic Carcinoma Project (not adrenocortical carcinoma) and mixes subtypes; salivary adenoid cystic carcinoma is ONCOTREE_CODE ACYC (935 of 1,049 samples). In ACYC, BCOR is mutated in 95 of 935 samples (10.2%).
Notes: A correct answer must: restrict every data point to salivary adenoid cystic carcinoma (OncoTree ACYC), and address BCOR somatic/germline data availability. Actionability needs OncoKB (not in the DB); classifying actionability from background knowledge is acceptable only if flagged as such. Must not: mix in lung, breast or other ACC-labelled cancers, confuse ACC with adrenocortical carcinoma, or present knowledge-based actionability as retrieved OncoKB data.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/94

#104 Out of scopeOut of scope · msk_chord_2024write me python code that can query the timeline files for msk-chord

Notes: A correct answer must: write code against the public cBioPortal REST API (e.g. /api/studies/{studyId}/clinical-events). Must not: write code that needs direct ClickHouse credentials.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/95

#105 DataStudy discovery · All Studieslist the portal studies for pediatric cancers that were published in the last 5 years

Reference answer (checked 2026-09-24): Pediatric studies with a citation year of 2021 or later: aml_stjude_2024, pancan_pdx_uthsa_2023, rms_msk_2023, nbl_msk_2023, msk_ch_ped_2021, pancan_mappyacts_2022, mixed_kunga_msk_2022, lgg_ctf_synodos_2025 (optionally pancan_ped_mai_msk_2025, which has no citation).
Notes: A correct answer must: list matching studies filtered by publication year (e.g. cancer_study.citation), each as a markdown link to its study summary page (https://www.cbioportal.org/study/summary?id=), covering most of the reference list. Must not: list bare study names without links, or include studies published before 2021.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/96

#106 DataStudy discovery · All StudiesWhich studies have RNA expression for renal cancer?

Notes: A correct answer must: list renal cancer studies with mRNA expression profiles, each linked to its study summary page. Must not: omit the study links.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/96

#107 DataStudy discovery · All StudiesWhat kind of cancer are there in the database?

Reference answer (checked 2026-09-24): The database has 119 cancer types (type_of_cancer) across its 548 studies.
Notes: A correct answer must: report the cancer types actually present in the database (e.g. distinct type_of_cancer from cancer_study joined to type_of_cancer: 119 types across 548 studies), grouped or summarised, with few tool calls. Must not: answer from the OncoTree ontology alone (search_oncotree lists all OncoTree types, not what is in the DB), base the answer on a sample of studies, or do extensive schema exploration for this simple question.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/97

#108 Out of scopeOut of scope · All StudiesCan you find a study that I may emulate that has a data_clinical_outcomes.txt file and associated meta file to interogate

Notes: A correct answer must: explain that the database stores loaded attributes, not source-file provenance, and point to cBioPortal/datahub or the file-format docs; be accurate that survival (KM) needs OS/DFS attributes, which can come from a separate clinical file. Must not: claim it can find this by querying the database, or overstate data-model rules.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/98

#109 DataVariants & hotspots · All StudiesI receive a report with the PIK3CA p.*1069Wext*3 mutation. Can you provide details of it

Notes: A correct answer must: report cBioPortal counts of the variant (and OncoKB annotation if available) and say that biological significance needs external sources (OncoKB, literature). Must not: imply it reviewed the scientific literature.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/99

#110 AnalysisAlteration frequency · All StudiesCan you explore the difference in mutation frequency between left-sided and right-sided CRC?

Notes: A correct answer must: explain that left vs right sidedness is defined by the splenic flexure (not colon vs rectum), use sub-site annotations where a study has them, or say the data doesn't support the comparison. Must not: substitute a colon-vs-rectum comparison.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/100

#111 NavigationCohort & clinical counts · All Studiescan you show me a study with longitudinal data and a patient that has multiple samples over time?

Notes: A correct answer must: pick a study with real longitudinal patient sampling (e.g. GLASS) and give at least one patient view link; explicitly retract anything it decides was wrong. Must not: cite multiple derived model samples (e.g. PDMR) as longitudinal patient data, or silently pivot to another study.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/101

#112 DataStudy discovery · All StudiesWhich cBioPortal studies include lung adenocarcinoma samples with mutation and copy-number data?

Notes: A correct answer must: return the complete list of studies with LUAD samples profiled for both mutations and CNA, with study links. Must not: stop mid-flow or return a truncated/unusable answer.
Source: https://github.com/cBioPortal/cbioportal-mcp/issues/102

#113 AnalysisExpression & multi-omics · lgg_tcga_pan_can_atlas_2018In lower grade glioma, are there genes which are overexpressed in any of the molecular subtypes?

Notes: A correct answer must: use an LGG study that has mRNA expression data (e.g. lgg_tcga_pan_can_atlas_2018) and link to the group comparison mRNA tab. (Rephrased from a follow-up in an LGG molecular-subtype conversation.) Must not: link to the mRNA tab of a study without expression data (e.g. lgggbm_tcga_pub).
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/6

#114 NavigationExpression & multi-omics · lgg_tcga_pan_can_atlas_2018Are there differences in chromosome arm 7p copy number between lower grade glioma molecular subtypes?

Notes: A correct answer must: return a valid link to the group comparison arm-level CNA view (or the copy-number tab), or explain what is available. (Rephrased from a follow-up.) Must not: return a broken URL.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/7

#115 NavigationExpression & multi-omics · lgg_tcga_pan_can_atlas_2018Are there DNA methylation differences between lower grade glioma molecular subtypes?

Notes: A correct answer must: link to the group comparison DNA methylation tab (probe-level, e.g. cg00000292). (Rephrased from follow-ups.) Must not: claim group comparison has no methylation tab.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/8

#116 NavigationVariants & hotspots · luad_tcga_pan_can_atlas_2018What are the frequencies of different KRAS mutations in TCGA PanCan Lung Adenocarcinoma?

Notes: A correct answer must: link directly to the results view Mutations tab for KRAS in luad_tcga_pan_can_atlas_2018. Must not: link to OncoPrint and explain how to navigate to Mutations, or add "what you'll find" summaries from background knowledge.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/10

#117 NavigationAlteration frequency · msk_impact_2017What cancer types in MSK-IMPACT have mutations in TP53?

Notes: The question does not pin a version, so msk_impact_2017 and msk_impact_50k_2026 are both acceptable. A correct answer must: give as primary link a study view with a TP53 mutation filter applied (cancer type breakdown), with the results view Cancer Types Summary as a secondary link. Must not: give only results-view links.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/11

#118 NavigationExpression & multi-omics · lgg_tcga_pan_can_atlas_2018In TCGA lower grade glioma, show me IDH1 mRNA expression by IDH1 mutation status.

Notes: A correct answer must: link to the Plots tab with axes pre-set (horizontal: IDH1 mutation type, vertical: IDH1 mRNA expression). (Rephrased from a follow-up.) Must not: tell the user to configure axes manually, or mention a "single-variable plot".
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/12

#119 AnalysisSurvival & outcomes · lgg_tcga_pan_can_atlas_2018In lower grade glioma, how do outcomes differ for IDH1 altered vs EGFR altered patients?

Notes: A correct answer must: treat "altered" as all alteration types and compare IDH1-altered vs EGFR-altered groups (e.g. results view with IDH1 and EGFR, comparison/survival tab with those groups selected). (Also covers cbioportal-navigator#14.) Must not: build groups from mutations only, or claim results view cannot compare gene-specific altered groups.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/13

#120 NavigationCo-occurrence & exclusivity · lgg_tcga_pan_can_atlas_2018is there a relationship between cic mutation and 19q del in lgg?

Notes: A correct answer must: link directly to the group comparison arm-level CNA subtab for CIC-mutant vs other samples, where 19q deletion is shown. Must not: tell the user to look for 19q genes manually in the enrichment table.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/15

#121 NavigationAlteration frequency · lgg_tcga_pan_can_atlas_2018In TCGA lower grade glioma, show me samples with EGFR gains.

Notes: A correct answer must: filter study view using a gene-specific EGFR CNA chart (GAIN). (Rephrased from a follow-up.) Must not: filter with the global CNA chart.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/16

#122 NavigationPatient & sample lookup · lgg_tcga_pan_can_atlas_2018In TCGA lower grade glioma, filter to samples that are both IDH1 and TP53 mutant and show me the summary page.

Notes: A correct answer must: produce a study view URL containing both gene filters (IDH1 AND TP53). (Rephrased from a follow-up.) Must not: drop one of the genes from the filter.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/19

#123 NavigationPatient & sample lookup · lgg_tcga_pan_can_atlas_2018In TCGA lower grade glioma, show me samples that are TP53 mutant or EGFR amplified.

Notes: A correct answer must: use an approach the UI supports (e.g. results view OQL `TP53: MUT; EGFR: AMP`, or custom selection) or clearly state the UI limitation. (Rephrased from a follow-up.) Must not: return a filter JSON the frontend can't display without saying so.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/20

#124 NavigationExpression & multi-omics · All Studiesshow me EGFR expression across cancer types

Notes: A correct answer must: link to the Plots tab with mRNA expression (vertical) vs CANCER_TYPE_DETAILED (horizontal). (Also covers cbioportal-navigator#29.) Must not: route to alteration frequencies / Cancer Types Summary, or set up CNA vs expression axes.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/21

#125 NavigationVariants & hotspots · All Studiesshow me point mutations in EGFR in lung cancer

Notes: A correct answer must: return a working link (after choosing or stating a study, e.g. luad_tcga_pan_can_atlas_2018), treating point mutations as all SNVs rather than only missense. Must not: return a blank response.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/22

#126 DataAlteration frequency · gbm_tcga_pan_can_atlas_2018are there cdkn2a het losses in gbm tcga study?

Reference answer (checked 2026-09-23): Yes. In gbm_tcga_pan_can_atlas_2018, 114 of 575 CNA-profiled samples (114 patients) have a CDKN2A shallow deletion (HETLOSS); 322 have a deep deletion (HOMDEL).
Notes: A correct answer must: use OQL `CDKN2A: HETLOSS` (or HETLOSS calls) and report the count separately from deep deletions. Must not: show all CDKN2A alterations as the answer.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/25

#127 NavigationPatient & sample lookup · gbm_tcga_pan_can_atlas_2018show me gbm with mgmt hypermethylation

Notes: A correct answer must: look up the available MGMT methylation probes, list them with gene/region annotations (e.g. "cg12434587 — MGMT, TSS..."), filter on a probe and mention the alternatives. (Also covers cbioportal-navigator#17.) Must not: tell the user to find probes manually, or list bare probe IDs.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/26

#128 DataStudy discovery · All StudiesIs there a lower grade glioma study with race data?

Expected links: https://www.cbioportal.org/study?id=lgg_tcga_pan_can_atlas_2018
Notes: A correct answer must: list matching studies with direct study view links. (Rephrased from a follow-up.) Must not: list study names without links.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/27

#129 AnalysisSurvival & outcomes · All Studiescompare atrx mutant vs cic mutant lgg - are there different outcomes?

Notes: A correct answer must: build groups from mutations only (OQL `ATRX: MUT`, `CIC: MUT`, or study view mutation filters) and link to comparison/survival. Must not: include all alteration types in the groups.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/28

#130 AnalysisSurvival & outcomes · All Studiesare there different outcomes for idh1 mutant vs egfr amp in lgg?

Notes: A correct answer must: compare IDH1 mutations only vs EGFR amplifications only. Must not: include all alteration types for either gene.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/28

#131 AnalysisCo-occurrence & exclusivity · All Studieswhat other genes are altered in kras mutant crc or luad

Notes: A correct answer must: compare KRAS-mutant vs wild-type with the wild-type group restricted to mutation-profiled samples (or use a results-view comparison). Must not: count unprofiled samples as wild-type.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/36

#132 NavigationAlteration frequency · All TCGA PancanCreate an OncoPrint with a merged track for the EGFR family genes (EGFR, ERBB2, ERBB3, ERBB4) across TCGA PanCancer Atlas studies.

Notes: A correct answer must: use valid OQL merged-track syntax (quoted label in brackets, e.g. ["EGFR FAMILY" EGFR ERBB2 ERBB3 ERBB4]) and check errors against the OQL reference. (Rephrased from the issue's description.) Must not: invent syntax fixes or limitations when an OQL error comes back.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/40

#133 DataAlteration frequency · msk_impact_50k_2026What are the EGFR mutation frequencies across cancer types in the MSK-IMPACT 50K study?

Notes: A correct answer must: use msk_impact_50k_2026 and name the study in the first sentence. (Rephrased from a follow-up.) Must not: silently substitute MSK-CHORD (msk_chord_2024).
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/41

#134 AnalysisAlteration frequency · All Studiescompare egfr mutations between lung and brain cancer

Notes: A correct answer must: build two groups each restricted to its own study/cancer type (e.g. LUAD vs GBM PanCancer) with distinct sample counts. Must not: produce groups that both contain the full combined cohort.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/42

#135 AnalysisSurvival & outcomes · All StudiesHow does overall survival compare between prostate cancer patients where TMPRSS2 is acting as the upstream donor to an ERG fusion event and patients without this fusion?

Notes: Group comparison links are session-based (`comparisonId` differs on every run): judge a comparison link by the groups and view the rendered page shows, not by matching the reference id or URL. A correct answer must: use fusion-specific OQL for TMPRSS2::ERG and hand survival to Kaplan-Meier via a cBioPortal link. Must not: fall back to "TMPRSS2 or ERG altered", or claim cBioPortal can't filter a specific fusion pair.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/43

#136 AnalysisCo-occurrence & exclusivity · brca_tcga_pan_can_atlas_2018In the TCGA Breast Cancer study, do TP53 mutations and high MYC expression co-occur or are they mutually exclusive?

Reference answer: They co-occur more than expected by chance (not mutually exclusive), per the user's review of the result.
Notes: A correct answer must: query mutations only (e.g. `TP53: MUT; MYC: EXP>2`) without adding GISTIC/SV profiles, and interpret co-occurrence from an actual test (cBioPortal mutual exclusivity tab). Must not: include all TP53 alterations, or conclude independence from the overlap percentage.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/43

#137 DataCohort & clinical counts · gbm_tcga_pan_can_atlas_2018In the TCGA Glioblastoma multiforme study (gbm_tcga_pan_can_atlas_2018), how many patients have an IDH1 R132H mutation and how many are IDH1 wild-type.

Reference answer (checked 2026-09-23): 22 patients have IDH1 R132H. 366 patients are IDH1 wild-type (no IDH1 mutation) out of 390 mutation-profiled patients; 2 more carry other IDH1 R132 variants (R132G, R132C), and 195 patients were not profiled for mutations.
Notes: A correct answer must: define wild-type as mutation-profiled patients with no IDH1 mutation of any kind. Must not: count unprofiled patients as wild-type, or define wild-type as "not R132H".
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/43

#138 NavigationVariants & hotspots · All Studiesshow me all KRAS mutations in colorectal cancer that are not at position 12

Notes: OQL allows one `!=` exclusion per gene: `KRAS: MUT != G12` excludes every codon-12 mutation. Position ranges also work: `KRAS: MUT = (-11) MUT = (13-)`. Invalid or wrong: `!=` combined with a range, two `!=` terms on one gene, `G12*` (excludes only the nonsense change), or a `p.` prefix. Judge by what the rendered page shows. A correct answer must: give a link whose OQL excludes codon-12 KRAS mutations in colorectal cancer (e.g. `KRAS: MUT != G12` or `KRAS: MUT = (-11) MUT = (13-)`). Must not: emit invalid OQL or show codon-12 mutations.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/44

#139 NavigationVariants & hotspots · All Studiesshow me cholangio with idh1 mutations other than r132

Notes: OQL allows one `!=` exclusion per gene: `IDH1: MUT != R132` excludes every R132 mutation. Position ranges also work: `IDH1: MUT = (-131) MUT = (133-)`. Invalid or wrong: `!=` combined with a range, two `!=` terms on one gene, `R132*` (excludes only the nonsense change), or a `p.` prefix. Judge by what the rendered page shows. A correct answer must: return links that include all IDH1 mutations except R132 in cholangiocarcinoma (e.g. `IDH1: MUT != R132`), or state which view can't express it; if no samples match, say so. Must not: filter to IDH1 missense/R132 (the opposite of the request).
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/44

#140 NavigationVariants & hotspots · All Studieswhat's the frequency of different idh1 mutations in glioma vs cholangio vs chndrosarcoma? Give me a link to compare the frequency of the specific IDH1 mutations in those cancer types.

Notes: A correct answer must: link to Group Comparison → Mutations tab with one group per cancer type, and describe that tab accurately. (Question joins the first turn with the follow-up.) Must not: link to overall IDH1 frequency (study view), or describe tab capabilities wrongly.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/45

#141 AnalysisExpression & multi-omics · All Studiesis there a relatinoship between mgmt methylation and idh1 mutation in glioma?

Notes: A correct answer must: choose a glioma study that has DNA methylation data before linking to Plots / group comparison. Must not: link to views for a study without methylation data.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/46

#142 NavigationAlteration frequency · All StudiesGive me an OncoPrint for RTK genes in lung cancer, limited to driver events.

Notes: A correct answer must: use the bare `DRIVER` OQL modifier. (Rephrased from a two-turn exchange.) Must not: spell out MUT_DRIVER/AMP_DRIVER per alteration type, or invent OQL syntax (e.g. a required "p." prefix).
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/47

#143 NavigationVariants & hotspots · All Studiesshow me P135L mutations in $p14^{ARF}$

Notes: A correct answer must: resolve p14ARF to CDKN2A (ARF isoform), keep the narrative and the link on the same cohort (pan-cancer), and say so if no samples match. Must not: say it will use TCGA PanCancer Atlas but link to a single breast study.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/48

#144 NavigationAlteration frequency · All Studiesare lung carcinosarcomas associated with BRIP1 mutations?

Notes: A correct answer must: keep the OncoPrint scoped to mutations since the question only asks about mutations. Must not: include copy-number tracks in the mutation-only OncoPrint.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/51

#145 AnalysisAlteration frequency · lusc_tcga_pan_can_atlas_2018, luad_tcga_pan_can_atlas_2018Which genes are enriched for mutations between NSCLC vs squamous cell carcinoma?

Notes: A correct answer must: reframe as LUAD vs LUSC and deliver a working group comparison link (mutation enrichment), or clearly explain why not. Must not: end with "unable" without a link or reason.
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/52

#146 NavigationTreatment · msk_chord_2024In the MSK-CHORD study, is it possible to see which patients received radiation therapy?

Notes: A correct answer must: use attributes/timeline data that actually exist in msk_chord_2024 and return a valid study view filter or timeline route. (Rephrased from a follow-up.) Must not: use a non-existent filter (e.g. filterAttributeId=PRIOR_MED_TO_MSK with Radiation Therapy).
Source: https://github.com/cBioPortal/cbioportal-navigator/issues/54

Multi-turn follow-ups

About 40% of real messages are follow-ups that need the earlier conversation. Each question is the user's next message after scripted turns: swapping a gene, study or cohort, breaking results down, accepting an offered option, answering a clarifying question, pushing back on a correct number, asking for significance or citations, and follow-ups in other languages.

Questions: input/questions-multiturn.yaml · 3 runs

8 Data2 Navigation2 Analysis2 Out of scope

1 Study discovery1 Cohort & clinical counts5 Alteration frequency2 Variants & hotspots1 Expression & multi-omics2 Survival & outcomes2 Out of scope

uv run cbioportal-mcp-qa run --questions-file input/questions-multiturn.yaml

#1001 DataVariants & hotspots · msk_impact_50k_2026 · follow-up (turn 2)How about G12C?

Conversation so far: User: What is the % of KRAS G12D across different tumor types? Use the MSK-IMPACT dataset.
Assistant: In MSK-IMPACT (msk_impact_50k_2026), KRAS G12D by cancer type (samples profiled for KRAS): - Pancreatic Cancer: 964 / 2,851 (33.8%) - Colorectal Cancer: 674 / 5,277 (12.8%) - Cancer of Unknown Primary: 100 / 1,581 (6.3%) - Endometrial Cancer: 135 / 2,445 (5.5%) - Non-Small Cell Lung Cancer: 285 / 7,867 (3.6%)
Reference answer (checked 2026-09-25): Non-Small Cell Lung Cancer 922 / 7,867 (11.7%) is highest; then Colorectal 158 / 5,277 (3.0%), Cancer of Unknown Primary 48 / 1,581 (3.0%), Endometrial 30 / 2,445 (1.2%), Pancreatic 33 / 2,851 (1.2%).
Notes: A correct answer must: report KRAS G12C by cancer type in the same study (msk_impact_50k_2026) with the same KRAS-profiled denominators, and show NSCLC as clearly the highest (~11.7%). Must not: switch to another study, or ask the user which gene/study they mean.

#1002 DataAlteration frequency · brca_tcga_pan_can_atlas_2018 · follow-up (turn 2)break that down by PAM50 subtype please

Conversation so far: User: How many samples in the TCGA PanCancer Atlas breast cancer study have PIK3CA mutations?
Assistant: In Breast Invasive Carcinoma (TCGA, PanCancer Atlas), 347 of 1,066 profiled samples (32.6%) have a PIK3CA mutation.
Reference answer (checked 2026-09-25): LumA 237 / 499 (47.5%), Her2 26 / 78 (33.3%), LumB 59 / 197 (29.9%), Normal-like 8 / 36 (22.2%), Basal 12 / 171 (7.0%).
Notes: A correct answer must: give PIK3CA mutation frequency per PAM50 subtype (the SUBTYPE attribute) in brca_tcga_pan_can_atlas_2018, highest in Luminal A (~47%) and lowest in Basal (~7%). Must not: use a different study or report subtype counts without the per-subtype denominator.

#1003 DataVariants & hotspots · lusc_tcga_pan_can_atlas_2018 · follow-up (turn 2)And in lung squamous?

Conversation so far: User: What are the most common KRAS mutations in TCGA lung adenocarcinoma?
Assistant: In Lung Adenocarcinoma (TCGA, PanCancer Atlas), 168 of 566 profiled samples (29.7%) have a KRAS mutation. The most common: - G12C: 70 (12.4%) - G12V: 40 (7.1%) - G12D: 20 (3.5%) - G12A: 17 (3.0%)
Reference answer (checked 2026-09-25): In lusc_tcga_pan_can_atlas_2018 only 7 of 484 samples (1.4%) have a KRAS mutation, each a different variant (G12A, G12V, G13C, Q61H, E3K, V14I, R123*); there is no G12C.
Notes: A correct answer must: use the TCGA PanCancer Atlas lung squamous study (lusc_tcga_pan_can_atlas_2018) and say KRAS mutations are rare there (~1.4%, 7 of 484) with no recurrent hotspot. Must not: report G12C (or any variant) as common in lung squamous, or reuse the adenocarcinoma numbers.

#1004 DataAlteration frequency · brca_tcga_pan_can_atlas_2018 · follow-up (turn 2)変異だけでなく、ホモ接合性欠失（deep deletion）も含めて計算してください。

Conversation so far: User: TCGA PanCancer Atlas の乳がんで TP53 変異の頻度を教えてください。
Assistant: Breast Invasive Carcinoma (TCGA, PanCancer Atlas) では、TP53 変異はプロファイルされた 1,066 サンプル中 347 サンプル（32.6%）に見られます。
Reference answer (checked 2026-09-25): About 35% — 351 of 996 samples profiled for both mutations and CNA (344 mutated, 9 with TP53 deep deletion).
Notes: A correct answer must: reply in Japanese, and report TP53 mutation OR deep deletion (HOMDEL) in brca_tcga_pan_can_atlas_2018 at roughly 33-35% (351 / 996 using the mutation+CNA-profiled samples; another denominator such as all samples, giving ~33%, is fine when stated), noting deep deletions add only a few samples. Must not: count shallow deletions or gains, or answer in English only.

#1005 DataAlteration frequency · luad_tcga_pan_can_atlas_2018 · follow-up (turn 2)那 KRAS 呢？

Conversation so far: User: 在TCGA肺腺癌研究中，EGFR突变的频率是多少？
Assistant: 在 Lung Adenocarcinoma (TCGA, PanCancer Atlas) 中，566 个检测样本中有 70 个（12.4%）携带 EGFR 突变。
Reference answer (checked 2026-09-25): 168 of 566 samples (29.7%).
Notes: A correct answer must: reply in Chinese, and give KRAS mutation frequency in luad_tcga_pan_can_atlas_2018: 168 / 566 (29.7%). Must not: switch to another study or answer in English only.

#1006 DataAlteration frequency · TCGA PanCancer Atlas (32 studies) · follow-up (turn 2)yes please

Conversation so far: User: How often is TP53 mutated in the TCGA PanCancer Atlas cohort?
Assistant: TP53 is the most frequently mutated gene across the TCGA PanCancer Atlas studies, but its rate varies a lot by cancer type. Would you like me to break it down by cancer type?
Reference answer (checked 2026-09-25): Highest: Ovarian Epithelial Tumor 373 / 523 (71.3%), Head and Neck 357 / 515 (69.3%), Non-Small Cell Lung Cancer 699 / 1,050 (66.6%), Esophagogastric 371 / 618 (60.0%), Pancreatic 107 / 179 (59.8%), Colorectal 314 / 534 (58.8%).
Notes: A correct answer must: give TP53 mutation frequency per cancer type across the TCGA PanCancer Atlas studies with per-type denominators, with ovarian, head and neck and NSCLC at the top (~67-71%). Must not: ask the user again what they want broken down, or sum mutation events across studies.

#1007 DataCohort & clinical counts · msk_chord_2024 · follow-up (turn 2)That seems low, I'm pretty sure MSK-CHORD has about 50,000 patients.

Conversation so far: User: How many patients are in MSK-CHORD?
Assistant: MSK-CHORD (msk_chord_2024) includes 24,950 patients with 25,040 samples.
Reference answer (checked 2026-09-25): 24,950 patients (25,040 samples).
Notes: A correct answer must: keep the correct count for msk_chord_2024 (24,950 patients), ideally after re-checking, and politely explain the difference; it may point out that the ~48,000-patient cohort is the separate MSK-IMPACT 50K study (msk_impact_50k_2026). Must not: agree that MSK-CHORD has ~50,000 patients or change the number to match the user.

#1008 DataStudy discovery · TARGET studies · follow-up (turn 2)TARGET, the pediatric one

Conversation so far: User: How can I find the TARGET information?
Assistant: Do you mean the NCI TARGET program (Therapeutically Applicable Research to Generate Effective Treatments — pediatric cancer studies), or something else, such as studies on a particular drug target?
Reference answer (checked 2026-09-25): TARGET studies in cBioPortal: alal_target_gdc, aml_target_gdc, bll_target_gdc, nbl_target_gdc, os_target_gdc, wt_target_gdc (TARGET GDC, 2025) and all_phase2_target_2018_pub, aml_target_2018_pub, nbl_target_2018_pub, rt_target_2018_pub, wt_target_2018_pub (TARGET, 2018).
Notes: A correct answer must: list the TARGET studies (most of the 11 above, covering leukemias, neuroblastoma, osteosarcoma, Wilms' and rhabdoid tumor) with a link to each study's cBioPortal page. Must not: include non-TARGET studies whose names merely contain "target" (e.g. biliary_tract_adc_targets_msk_2026, sarcoma_msk_2026), or ask again which TARGET the user means.

#1009 NavigationAlteration frequency · luad_tcga_pan_can_atlas_2018 · follow-up (turn 2)can you do the same for lung adenocarcinoma?

Conversation so far: User: Show me an OncoPrint of KRAS, TP53 and CDKN2A in TCGA pancreatic cancer.
Assistant: Here is the OncoPrint for KRAS, TP53 and CDKN2A in Pancreatic Adenocarcinoma (TCGA, PanCancer Atlas): [Open OncoPrint](https://www.cbioportal.org/results/oncoprint?cancer_study_list=paad_tcga_pan_can_atlas_2018&gene_list=KRAS%20TP53%20CDKN2A)
Expected links: https://www.cbioportal.org/results/oncoprint?cancer_study_list=luad_tcga_pan_can_atlas_2018&gene_list=KRAS%20TP53%20CDKN2A
Notes: A correct answer must: give a results-view OncoPrint link for luad_tcga_pan_can_atlas_2018 with the same three genes (KRAS, TP53, CDKN2A). Must not: drop or change genes, keep the pancreatic study, or ask which genes to use.

#1010 NavigationSurvival & outcomes · lgg_tcga_pan_can_atlas_2018 · follow-up (turn 2)option 1 please

Conversation so far: User: I'm interested in IDH1 in lower grade glioma. What can we look at?
Assistant: In Brain Lower Grade Glioma (TCGA, PanCancer Atlas), IDH1 is mutated in most tumors. A few directions: 1. Compare overall survival of IDH1-mutant vs IDH1-wild-type patients 2. See which genes are co-mutated with IDH1 (e.g. TP53, ATRX, CIC) 3. Look at IDH1 mutation positions (R132H and others) in the mutation diagram Which would you like?
Notes: A correct answer must: give a cBioPortal group comparison (or results-view survival) link for lgg_tcga_pan_can_atlas_2018 comparing IDH1-mutant vs IDH1-wild-type patients, pointing to the Survival tab for the Kaplan-Meier curves and log-rank p-value. Must not: invent median survival times, hazard ratios or p-values, or ask again which option the user wants.

#1011 AnalysisSurvival & outcomes · coadread_tcga_pan_can_atlas_2018 · follow-up (turn 2)is the difference in survival significant?

Conversation so far: User: In TCGA colorectal cancer, do patients with high PTK2 expression have different survival from patients with low PTK2 expression?
Assistant: In Colorectal Adenocarcinoma (TCGA, PanCancer Atlas), the Kaplan-Meier curves for patients with PTK2 mRNA above vs below the median look slightly apart, with the high-PTK2 group doing a little worse early on. You can build this comparison in cBioPortal by querying PTK2 with an expression threshold (e.g. `PTK2: EXP>0`) and opening the Comparison → Survival tab.
Notes: A correct answer must: say it cannot compute a significance test itself and point the user to the log-rank p-value on the Survival tab of the comparison link (or give one only if it comes from a tool that computed it). Must not: state a p-value, hazard ratio or "significant/not significant" conclusion it did not compute.

#1012 Out of scopeOut of scope · All Studies · follow-up (turn 2)what literature? citations please

Conversation so far: User: How are PTK2 and STING1 expression related in colorectal adenocarcinoma?
Assistant: In Colorectal Adenocarcinoma (TCGA, PanCancer Atlas), PTK2 and STING1 mRNA expression show only a weak correlation. Tumors with active FAK (PTK2) signaling are often reported in the literature to suppress innate immune signaling, which could relate to STING pathway activity.
Notes: A correct answer must: acknowledge that the literature remark was general background it cannot back with verified citations from cBioPortal, and either cite only sources it can verify (e.g. publications of cBioPortal studies) or suggest searching PubMed. Must not: produce specific paper titles, authors, journals, years, DOIs or PMIDs that it did not retrieve from a tool.

#1013 Out of scopeOut of scope · All Studies · follow-up (turn 2)what if the genes, like AC002511.1 or LOC100506271, just aren't in HGNC or NCBI at all?

Conversation so far: User: I'm importing a study into my own cBioPortal instance and the validator says some genes are not known to cBioPortal. What does that mean?
Assistant: The validator checks every Hugo_Symbol / Entrez_Gene_Id in your data files against the gene table loaded in your instance (built from HGNC/NCBI). "Gene not known" means the symbol or ID isn't in that table — usually an outdated symbol, an alias, or a typo.
Notes: A correct answer must: explain that genes absent from the instance's gene table can't be loaded (the importer skips those rows; the validator warns), and suggest mapping them to a current HGNC symbol / Entrez ID where one exists or accepting that they will be left out. Must not: invent validator/importer flags or config options, or claim cBioPortal can load arbitrary unknown gene symbols as-is.

#1014 AnalysisExpression & multi-omics · brca_metabric · follow-up (turn 2)Can you do this with METABRIC instead?

Conversation so far: User: How does ESR1 mRNA expression differ between PAM50 subtypes in TCGA breast cancer?
Assistant: In Breast Invasive Carcinoma (TCGA, PanCancer Atlas), ESR1 mRNA is highest in Luminal A and Luminal B tumors, intermediate in Normal-like and HER2-enriched, and lowest in Basal-like tumors. You can see the distribution here: [ESR1 expression by subtype](https://www.cbioportal.org/results/plots?cancer_study_list=brca_tcga_pan_can_atlas_2018&gene_list=ESR1)
Notes: A correct answer must: repeat the ESR1 mRNA-by-PAM50-subtype comparison in brca_metabric (its Pam50 + Claudin-low subtype attribute), reaching the same pattern (Luminal highest, Basal lowest), with a link to the METABRIC study or plots view. Must not: stay on the TCGA study, or claim a statistically significant difference (p-value) without a test having been run.

Subset questions

The main set leans on per-study statistics (top genes in a study, a gene's frequency in a study), which precomputed tables and per-study tools answer directly. Here the cohort is a clinical subset of a study (sample type, sex, smoking, stage, MSI, receptor status, OncoTree code) or a pooled set of studies, so the agent must write the SQL and must not answer with the whole-study number. Covers subset frequencies, pooled studies, subset comparisons, co-occurrence, top genes, patient vs sample level, clinical counts in mutation-defined groups and multi-turn narrowing.

Questions: input/questions-subset.yaml · 1 run

19 Data7 Analysis

2 Cohort & clinical counts17 Alteration frequency1 Variants & hotspots3 Co-occurrence & exclusivity1 Survival & outcomes2 Patient & sample lookup

uv run cbioportal-mcp-qa run --questions-file input/questions-subset.yaml

#2001 DataAlteration frequency · msk_chord_2024In MSK-CHORD, what percentage of non-small cell lung cancer samples from never-smokers have an EGFR mutation?

Reference answer (checked 2026-10-01): 49.5% (1,003 of 2,027 NSCLC samples from patients with SMOKING_PREDICTIONS_3_CLASSES = Never, all profiled for EGFR).
Notes: Subtype: clinical-subset frequency. Subset: CANCER_TYPE = Non-Small Cell Lung Cancer (sample) and SMOKING_PREDICTIONS_3_CLASSES = Never (patient). Denominator: subset samples profiled for EGFR. A correct answer must: restrict to never-smokers within NSCLC and report about 49.5% (1,003 / 2,027). Must not: report the all-NSCLC EGFR frequency (24.2%, 1,890 / 7,809) from the per-cancer-type table as the answer.

#2002 DataAlteration frequency · msk_chord_2024How often is PIK3CA mutated in HR-positive, HER2-negative breast cancer samples in the MSK-CHORD Study?

Reference answer (checked 2026-10-01): 40.8% of samples (1,434 of 3,515 breast cancer samples from HR+ / HER2- patients, all profiled for PIK3CA).
Notes: Subtype: clinical-subset frequency. Subset: CANCER_TYPE = Breast Cancer (sample), HR = Yes and HER2 = No (patient attributes). Denominator: subset samples profiled for PIK3CA. A correct answer must: use the HR and HER2 attributes to select HR+/HER2- breast cancer and report about 40.8% (1,434 / 3,515). Must not: report the all-breast-cancer PIK3CA frequency (35.5%, 1,908 / 5,368) as the answer.

#2003 DataAlteration frequency · msk_impact_2017In the 2017 MSK-IMPACT study (msk_impact_2017), what fraction of metastatic samples carry a KRAS mutation, across all cancer types?

Reference answer (checked 2026-10-01): 14.7% (696 of 4,732 metastasis samples, all profiled for KRAS).
Notes: Subtype: clinical-subset frequency. Subset: SAMPLE_TYPE = Metastasis. Denominator: metastatic samples profiled for KRAS. A correct answer must: restrict to SAMPLE_TYPE = Metastasis and report about 14.7% (696 / 4,732). Must not: report the whole-study KRAS frequency (15.0%, 1,643 / 10,945) as the answer.

#2004 DataAlteration frequency · msk_chord_2024What is the BRAF mutation frequency among MSI-high (microsatellite instable) colorectal cancer samples in MSK-CHORD?

Reference answer (checked 2026-10-01): 42.2% (247 of 586 colorectal samples with MSI_TYPE = Instable, all profiled for BRAF).
Notes: Subtype: clinical-subset frequency. Subset: CANCER_TYPE = Colorectal Cancer and MSI_TYPE = Instable (sample attributes). Denominator: subset samples profiled for BRAF. A correct answer must: select MSI-instable colorectal samples via MSI_TYPE and report about 42.2% (247 / 586). Must not: report the all-colorectal BRAF frequency (11.1%, 618 / 5,543) as the answer, or define MSI-high by an invented MSI_SCORE cutoff without saying so.

#2005 DataAlteration frequency · msk_chord_2024Among metastatic pancreatic cancer samples in MSK-CHORD, how many have a SMAD4 mutation, and what percentage is that?

Reference answer (checked 2026-10-01): 225 of 1,124 metastatic pancreatic cancer samples (20.0%).
Notes: Subtype: clinical-subset frequency. Subset: CANCER_TYPE = Pancreatic Cancer and SAMPLE_TYPE = Metastasis. Denominator: subset samples profiled for SMAD4. A correct answer must: give 225 mutated of 1,124 metastatic pancreatic samples (about 20.0%). Must not: report the all-pancreatic numbers (604 / 3,109, 19.4%) from the per-cancer-type table; the percentages are close, so the counts and denominator decide.

#2006 DataAlteration frequency · coadread_tcga_pan_can_atlas_2018, paad_tcga_pan_can_atlas_2018, luad_tcga_pan_can_atlas_2018Pooling the TCGA PanCancer Atlas colorectal, pancreatic and lung adenocarcinoma studies into one cohort, what percentage of samples have a KRAS mutation?

Reference answer (checked 2026-10-01): 39.3% pooled (503 of 1,279 KRAS-profiled samples). Per study for context: colorectal 218 / 534 (40.8%), pancreatic 117 / 179 (65.4%), lung adenocarcinoma 168 / 566 (29.7%).
Notes: Subtype: pooled studies. Studies: coadread_tcga_pan_can_atlas_2018, paad_tcga_pan_can_atlas_2018, luad_tcga_pan_can_atlas_2018. Denominator: samples profiled for KRAS across the three studies (1,279 of 1,344 samples). A correct answer must: give the pooled frequency of about 39.3% (503 / 1,279); per-study numbers are optional. Must not: give the unweighted mean of the per-study percentages (~45.3%) as the pooled frequency, or use all 1,344 samples (including unprofiled) as the denominator.

#2007 DataAlteration frequency · lgg_tcga_pan_can_atlas_2018, gbm_tcga_pan_can_atlas_2018If I combine the TCGA PanCancer Atlas lower grade glioma and glioblastoma studies, what is the overall IDH1 mutation frequency across their samples?

Reference answer (checked 2026-10-01): 46.1% (420 of 911 IDH1-profiled samples). Per study for context: LGG 395 / 514 (76.8%), GBM 25 / 397 (6.3%).
Notes: Subtype: pooled studies. Studies: lgg_tcga_pan_can_atlas_2018, gbm_tcga_pan_can_atlas_2018. Denominator: samples profiled for IDH1 (911 of 1,106 samples). A correct answer must: give the pooled frequency of about 46.1% (420 / 911). Must not: average the two study percentages (~41.6%), or divide by all 1,106 samples including those not profiled for mutations.

#2008 DataAlteration frequency · brca_tcga_pan_can_atlas_2018, ucec_tcga_pan_can_atlas_2018, cesc_tcga_pan_can_atlas_2018Across the TCGA PanCancer Atlas breast, endometrial and cervical cancer studies taken together, how many samples have a PIK3CA mutation and what is the pooled frequency?

Reference answer (checked 2026-10-01): 689 of 1,874 PIK3CA-profiled samples (36.8%). Per study for context: breast 347 / 1,066 (32.6%), endometrial 259 / 517 (50.1%), cervical 83 / 291 (28.5%).
Notes: Subtype: pooled studies. Studies: brca_tcga_pan_can_atlas_2018, ucec_tcga_pan_can_atlas_2018, cesc_tcga_pan_can_atlas_2018. Denominator: samples profiled for PIK3CA (1,874 of 1,910 samples). A correct answer must: give the pooled count and frequency, 689 / 1,874 (about 36.8%). Must not: average the per-study percentages (~37.1%) instead of pooling counts, or use the 1,910 total samples as the denominator.

#2009 AnalysisAlteration frequency · msk_chord_2024In MSK-CHORD breast cancer, is ESR1 mutated more often in metastatic samples than in primary tumors?

Reference answer (checked 2026-10-01): Yes, much more often: 11.0% of metastasis samples (271 / 2,464) vs 1.6% of primary samples (47 / 2,859).
Notes: Subtype: subset comparison. Subset: CANCER_TYPE = Breast Cancer, split by SAMPLE_TYPE (Primary vs Metastasis; 30 Local Recurrence and 15 Unknown samples are outside both groups). Denominator: samples profiled for ESR1 in each group. A correct answer must: compare the two groups with their own denominators (about 11.0% vs 1.6%) and conclude ESR1 mutations are enriched in metastases (consistent with endocrine-therapy resistance). Must not: report only the all-breast-cancer ESR1 frequency (6.0%, 323 / 5,368), or invent a p-value it did not compute.

#2010 AnalysisAlteration frequency · msk_chord_2024Compare the KRAS mutation rate in primary versus metastatic non-small cell lung cancer samples in MSK-CHORD.

Reference answer (checked 2026-10-01): Primary 29.0% (1,424 / 4,908) vs metastasis 26.3% (727 / 2,763); slightly higher in primary samples.
Notes: Subtype: subset comparison. Subset: CANCER_TYPE = Non-Small Cell Lung Cancer, split by SAMPLE_TYPE (103 Unknown and 35 Local Recurrence samples are outside both groups). Denominator: samples profiled for KRAS in each group. A correct answer must: report both group frequencies with denominators (about 29.0% vs 26.3%) and describe the difference as modest. Must not: report only the all-NSCLC KRAS frequency (28.0%, 2,183 / 7,809), or claim significance without a test.

#2011 AnalysisAlteration frequency · msk_chord_2024In MSK-CHORD, how does the EGFR mutation frequency in non-small cell lung cancer samples differ between women and men?

Reference answer (checked 2026-10-01): Women 28.0% (1,281 / 4,567 samples) vs men 18.8% (608 / 3,241 samples); EGFR mutations are more frequent in women.
Notes: Subtype: subset comparison. Subset: CANCER_TYPE = Non-Small Cell Lung Cancer (sample), split by the patient attribute GENDER (one NSCLC sample has GENDER = Unknown). Denominator: samples profiled for EGFR in each group. A correct answer must: give both frequencies with denominators (about 28.0% vs 18.8%) and say EGFR is more frequent in women. Must not: report only the all-NSCLC EGFR frequency (24.2%, 1,890 / 7,809).

#2012 AnalysisAlteration frequency · msk_impact_2017In msk_impact_2017, compare the TP53 mutation frequency in non-small cell lung cancer samples from previous/current smokers versus never-smokers.

Reference answer (checked 2026-10-01): Previous/current smokers 59.0% (608 / 1,031) vs never-smokers 44.7% (165 / 369); TP53 is more frequently mutated in smokers.
Notes: Subtype: subset comparison. Subset: CANCER_TYPE = Non-Small Cell Lung Cancer (sample), split by the patient attribute SMOKING_HISTORY (Prev/Curr Smoker vs Never; 266 Unknown and 2 blank samples are outside both groups). Denominator: samples profiled for TP53 in each group. A correct answer must: give both frequencies with denominators (about 59.0% vs 44.7%) and say TP53 is more frequent in smokers. Must not: report only the all-NSCLC TP53 frequency (55.0%, 917 / 1,668); give patient-level figures as the answer (smokers 581 / 972 patients, 59.8%; never-smokers 150 / 334, 44.9%); or fold Unknown smoking status (142 / 266) into either group.

#2013 AnalysisCo-occurrence & exclusivity · msk_chord_2024In non-small cell lung cancer samples from MSK-CHORD, do KRAS and STK11 mutations tend to co-occur? Give me the counts.

Reference answer (checked 2026-10-01): Yes, they co-occur. Of 7,809 NSCLC samples profiled for both: 529 have both, 1,654 KRAS only, 484 STK11 only, 5,142 neither (odds ratio about 3.4).
Notes: Subtype: co-occurrence in a subset. Subset: CANCER_TYPE = Non-Small Cell Lung Cancer; population: samples profiled for both genes. A correct answer must: give the overlap counts (529 both; 1,654 / 484 single-mutant) within NSCLC and conclude co-occurrence; an odds ratio is optional. Must not: compute the overlap over the whole study (all cancer types: 587 both, 6,541 KRAS only, 620 STK11 only of 25,040, OR about 2.5), or invent a p-value it did not compute; for significance it may point to cBioPortal's Mutual Exclusivity tab.

#2014 AnalysisCo-occurrence & exclusivity · msk_impact_2017Are KRAS and EGFR mutations mutually exclusive in metastatic non-small cell lung cancer samples from msk_impact_2017? How many samples have both?

Reference answer (checked 2026-10-01): They are largely mutually exclusive: of 672 metastatic NSCLC samples profiled for both, only 11 have both, 149 KRAS only, 156 EGFR only, 356 neither (odds ratio about 0.17).
Notes: Subtype: co-occurrence in a subset. Subset: CANCER_TYPE = Non-Small Cell Lung Cancer and SAMPLE_TYPE = Metastasis; population: samples profiled for both genes. A correct answer must: report 11 samples with both and describe the tendency toward mutual exclusivity, with the single-mutant counts. Must not: use all NSCLC samples (13 with both of 1,668; 436 KRAS only, 395 EGFR only, OR about 0.06), or invent a p-value it did not compute.

#2015 AnalysisCo-occurrence & exclusivity · msk_chord_2024In HR-positive, HER2-negative breast cancer samples in MSK-CHORD, how often do TP53 and PIK3CA mutations occur together, and do they tend to be exclusive?

Reference answer (checked 2026-10-01): Of 3,515 HR+/HER2- breast cancer samples profiled for both: 307 have both, 609 TP53 only, 1,127 PIK3CA only, 1,472 neither. Odds ratio about 0.66: they co-occur less often than expected, a tendency toward mutual exclusivity.
Notes: Subtype: co-occurrence in a subset. Subset: CANCER_TYPE = Breast Cancer (sample), HR = Yes and HER2 = No (patient); population: samples profiled for both genes. A correct answer must: give 307 samples with both mutations (8.7%) within HR+/HER2- breast cancer and describe the tendency toward exclusivity. Must not: report the all-breast-cancer numbers (597 with both of 5,368; 1,541 TP53 only, 1,311 PIK3CA only, OR about 0.57) as the answer, or invent a p-value it did not compute.

#2016 DataAlteration frequency · msk_chord_2024What are the five most frequently mutated genes in metastatic prostate cancer samples in MSK-CHORD?

Reference answer (checked 2026-10-01): TP53 33.6% (360), FOXA1 13.3% (142), SPOP 12.2% (131), APC 9.9% (106), PTEN 8.4% (90), of 1,070 metastatic prostate samples.
Notes: Subtype: top genes in a subset, ranked by frequency (ranking by count gives the same top 5). Subset: CANCER_TYPE = Prostate Cancer and SAMPLE_TYPE = Metastasis (1,070 samples). Per-gene denominator: subset samples profiled for the gene (all 1,070 for these genes). Next: KMT2D 7.9%, AR 7.3%. A correct answer must: list TP53, FOXA1, SPOP, APC, PTEN in that order with approximately these frequencies. Must not: give the whole-study top genes (TP53 52.4%, KRAS 28.5%, APC 19.1%, PIK3CA 14.8%, EGFR 8.6% of 25,040) or the all-prostate list (TP53 26.1%, FOXA1 14.8%, SPOP 13.0%, PTEN 7.1%, APC 6.4% of 3,211 samples).

#2017 DataAlteration frequency · msk_impact_2017Which genes are most frequently mutated in non-small cell lung cancer samples from never-smokers in the msk_impact_2017 study? Top 5 please.

Reference answer (checked 2026-10-01): EGFR 47.4% (175), TP53 44.7% (165), PIK3CA 8.1% (30), KRAS 7.0% (26), SETD2 6.2% (23), of 369 samples.
Notes: Subtype: top genes in a subset, ranked by frequency (ranking by count gives the same top 5). Subset: CANCER_TYPE = Non-Small Cell Lung Cancer (sample) and SMOKING_HISTORY = Never (patient); 369 samples, all profiled for these genes. Next: RB1 5.4%, ERBB2 5.1%. A correct answer must: rank EGFR first and TP53 second, followed by PIK3CA, KRAS and SETD2, with approximately these frequencies. Must not: give the all-NSCLC ranking (TP53 55.0%, KRAS 26.9%, EGFR 24.5%, STK11 15.9%, KEAP1 15.0% of 1,668 samples) or the whole-study one (TP53 41.5%, KRAS 15.0%, TERT 13.3%, PIK3CA 12.4%, APC 10.2% of 10,945), or patient-level figures as sample frequencies (EGFR 157 / 334 patients 47.0%, TP53 44.9%, PIK3CA 8.1%, KRAS 7.5%, SETD2 6.0%).

#2018 DataAlteration frequency · msk_chord_2024For microsatellite-stable colorectal cancer samples from patients with stage 4 disease in MSK-CHORD, what are the top 5 mutated genes?

Reference answer (checked 2026-10-01): TP53 78.7% (1,714), APC 78.6% (1,711), KRAS 46.6% (1,015), PIK3CA 18.6% (404), SMAD4 15.4% (335), of 2,177 samples.
Notes: Subtype: top genes in a subset, ranked by frequency (ranking by count gives the same top 5). Subset: CANCER_TYPE = Colorectal Cancer and MSI_TYPE = Stable (sample) and STAGE_HIGHEST_RECORDED = Stage 4 (patient); 2,177 samples. Per-gene denominator: subset samples profiled for the gene (2,177 for the top 5; TCF7L2 is next at 11.3% of 2,057). A correct answer must: list TP53, APC, KRAS, PIK3CA, SMAD4 with approximately these frequencies; TP53 and APC are effectively tied, so either order is fine. Must not: give the all-colorectal list (APC 74.7%, TP53 73.4%, KRAS 42.5%, PIK3CA 20.6%, FBXW7 15.7% of 5,543 samples) or the whole-study one (TP53 52.4%, KRAS 28.5%, APC 19.1%, PIK3CA 14.8%, EGFR 8.6%).

#2019 DataPatient & sample lookup · msk_impact_2017In msk_impact_2017, how many non-small cell lung cancer patients have an EGFR mutation, and how does that compare with the number of samples?

Reference answer (checked 2026-10-01): 370 of 1,567 NSCLC patients (23.6%) have an EGFR mutation in at least one NSCLC sample, versus 408 of 1,668 NSCLC samples (24.5%): some patients have several samples.
Notes: Subtype: patient vs sample level. Subset: CANCER_TYPE = Non-Small Cell Lung Cancer samples profiled for EGFR; patient level = patients with at least one such sample. A correct answer must: give the patient count (370 of 1,567) and the sample count (408 of 1,668) and explain the difference (patients with multiple samples). Must not: present the sample count (408) as the number of patients.

#2020 DataPatient & sample lookup · msk_impact_2017For metastatic breast cancer samples in msk_impact_2017, give the TP53 mutation frequency both per sample and per patient.

Reference answer (checked 2026-10-01): Per sample: 353 of 837 metastatic breast samples (42.2%). Per patient: 333 of 785 patients (42.4%).
Notes: Subtype: patient vs sample level. Subset: CANCER_TYPE = Breast Cancer and SAMPLE_TYPE = Metastasis, profiled for TP53; patient level = patients with at least one such sample, mutated if any of them has a TP53 mutation. A correct answer must: give both levels with their own denominators (353 / 837 samples and 333 / 785 patients). Must not: report the all-breast-cancer TP53 frequency (41.4%, 553 / 1,337), or divide patient counts by sample counts.

#2021 DataSurvival & outcomes · msk_chord_2024How many non-small cell lung cancer patients in MSK-CHORD have a KRAS G12C mutation, and how many of them are recorded as deceased?

Reference answer (checked 2026-10-01): 916 NSCLC patients have KRAS G12C; 494 of them (53.9%) are deceased (OS_STATUS = 1:DECEASED) and 422 living.
Notes: Subtype: clinical counts in a mutation-defined subset. Subset: patients with at least one NSCLC sample (profiled for KRAS) carrying KRAS G12C; OS_STATUS is a patient attribute. Context: among the 6,893 profiled NSCLC patients without G12C, 3,481 (50.5%) are deceased. A correct answer must: give 916 G12C patients and 494 deceased. Must not: count all KRAS-mutant NSCLC patients (2,183, of whom 1,121 deceased, 51.4%), or present the deceased share as a survival rate or median survival.

#2022 DataCohort & clinical counts · msk_chord_2024Among MSK-CHORD lung cancer (NSCLC) patients with an STK11 mutation, how many had stage 4 disease?

Reference answer (checked 2026-10-01): 426 of 1,013 STK11-mutant NSCLC patients (42.1%) have STAGE_HIGHEST_RECORDED = Stage 4; 587 are Stage 1-3.
Notes: Subtype: clinical counts in a mutation-defined subset. Subset: patients with at least one NSCLC sample (profiled for STK11) with an STK11 mutation; STAGE_HIGHEST_RECORDED is a patient attribute. Context: 2,995 of 6,796 STK11-wild-type profiled NSCLC patients (44.1%) are Stage 4. A correct answer must: give 426 stage 4 patients out of 1,013 STK11-mutant NSCLC patients. Must not: include STK11-mutant patients with other cancer types (all cancer types: 484 of 1,206 stage 4, 40.1%).

#2023 DataCohort & clinical counts · msk_chord_2024In MSK-CHORD, among breast cancer patients with a known liver-metastasis status, what share of those with a TP53 mutation have had liver metastases, compared with those without a TP53 mutation?

Reference answer (checked 2026-10-01): TP53-mutant: 1,000 of 2,048 patients (48.8%) have LIVER = Yes. TP53 not mutated: 1,322 of 2,969 patients (44.5%).
Notes: Subtype: clinical counts in a mutation-defined subset. Subset: patients with at least one breast cancer sample profiled for TP53 (all breast cancer patients are profiled); mutated = any such sample with a TP53 mutation. LIVER is a patient attribute; the denominators are patients with LIVER = Yes or No (90 and 261 patients with LIVER = Unknown are left out, as the question says). A correct answer must: use the LIVER metastatic-site attribute and give about 48.8% (1,000 / 2,048) vs 44.5% (1,322 / 2,969). Must not: keep LIVER = Unknown in the denominators (46.8%, 1,000 / 2,138 vs 40.9%, 1,322 / 3,230); report the all-breast-cancer rate without the TP53 split (2,322 / 5,017, 46.3%); or claim significance without a test.

#2024 DataAlteration frequency · msk_chord_2024 · follow-up (turn 2)Now only for metastatic samples from female patients.

Conversation so far: User: What are the top 5 most frequently mutated genes in the MSK-CHORD Study?
Assistant: In MSK-CHORD (msk_chord_2024), of 25,040 profiled samples: TP53 52.4%, KRAS 28.5%, APC 19.1%, PIK3CA 14.8%, EGFR 8.6%.
Reference answer (checked 2026-10-01): TP53 53.0% (2,782), KRAS 23.7% (1,244), PIK3CA 20.4% (1,072), APC 12.4% (649), EGFR 9.9% (518), of 5,247 metastatic samples from female patients.
Notes: Subtype: multi-turn narrowing. Subset: SAMPLE_TYPE = Metastasis (sample) and GENDER = Female (patient) in msk_chord_2024; 5,247 samples, all profiled for these genes. Ranked by frequency (ranking by count gives the same top 5). A correct answer must: stay in msk_chord_2024, apply both filters, and give this ranking (PIK3CA now ahead of APC) with approximately these frequencies. Must not: repeat the whole-study list (in the history), apply only one of the two filters (metastatic only: TP53 55.1%, KRAS 26.2%, APC 15.8%, PIK3CA 14.9%, EGFR 8.8% of 8,878; female only: TP53 51.5%, KRAS 27.6%, PIK3CA 19.8%, APC 15.1%, EGFR 10.3% of 13,923), or switch study.

#2025 DataAlteration frequency · msk_chord_2024 · follow-up (turn 2)How does that split between never-smokers and former or current smokers?

Conversation so far: User: What is the KRAS mutation frequency in non-small cell lung cancer in MSK-CHORD?
Assistant: In MSK-CHORD (msk_chord_2024), 2,183 of 7,809 profiled non-small cell lung cancer samples (28.0%) have a KRAS mutation.
Reference answer (checked 2026-10-01): Former/current smokers 35.7% (1,940 / 5,438) vs never-smokers 7.5% (152 / 2,027); 344 samples have Unknown smoking status (26.5%).
Notes: Subtype: multi-turn narrowing. Subset: CANCER_TYPE = Non-Small Cell Lung Cancer split by the patient attribute SMOKING_PREDICTIONS_3_CLASSES in msk_chord_2024. Denominator: samples profiled for KRAS in each group. A correct answer must: give both group frequencies with denominators (about 35.7% vs 7.5%) and note KRAS is far more common in smokers. Must not: switch study, or repeat the overall 28.0%.

#2026 DataVariants & hotspots · msk_chord_2024What percentage of lung adenocarcinoma samples in MSK-CHORD have KRAS G12C?

Reference answer (checked 2026-10-01): 13.3% (793 of 5,957 lung adenocarcinoma samples, ONCOTREE_CODE = LUAD).
Notes: Subtype: fast-path trap. Subset: ONCOTREE_CODE = LUAD (CANCER_TYPE_DETAILED Lung Adenocarcinoma); the per-cancer-type tables group by CANCER_TYPE, where LUAD is folded into Non-Small Cell Lung Cancer (7,809 samples, 916 with G12C, 11.7%). Denominator: LUAD samples profiled for KRAS. A correct answer must: restrict to lung adenocarcinoma and report about 13.3% (793 / 5,957). Must not: report the all-NSCLC G12C frequency (11.7%, 916 / 7,809) as the lung adenocarcinoma number.