Datasets:

mibohl
/

Bulk-RNA-Seq-Benchmark

task_id stringlengths 13 33	task_type stringclasses 2 values	TabPFN R/AUROC float64 0.34 0.94 ⌀	TabPFN rho float64 0.13 0.8 ⌀	TabFM R/AUROC float64 0.15 0.98	TabFM rho float64 0.19 0.82 ⌀	ElasticNet R/AUROC float64 0.15 0.94	ElasticNet rho float64 0.16 0.79 ⌀	RandomForest R/AUROC float64 0.13 0.93 ⌀	RandomForest rho float64 0.16 0.78 ⌀	MyModel R/AUROC float64 0.23 0.94	MyModel rho float64 0.25 0.82 ⌀
CPTAC::ajcc_stage_4	cls	0.787438	null	0.845108	null	0.757612	null	0.783043	null	0.724674	null
GSE107422::systemic_recurrence	cls	0.660317	null	0.652381	null	0.742857	null	0.61746	null	0.67619	null
GSE107995::tb_group	cls	0.827157	null	0.852106	null	0.837928	null	0.777426	null	0.841268	null
GSE109142::week4_remission	cls	0.640553	null	0.627976	null	0.662888	null	0.65491	null	0.648991	null
GSE112927::graft_loss	cls	0.522764	null	0.628862	null	0.547154	null	0.476016	null	0.634959	null
GSE116324::bortezomib_response	cls	0.615079	null	0.666667	null	0.719048	null	0.57381	null	0.663492	null
GSE120622::relapse	cls	0.729911	null	0.717361	null	0.717212	null	0.695412	null	0.780109	null
GSE154261::recurrence	cls	0.626455	null	0.648016	null	0.635384	null	0.573049	null	0.635648	null
GSE156699::chemo_response	cls	0.673571	null	0.691071	null	0.695357	null	0.678393	null	0.694643	null
GSE156902::group	cls	0.941454	null	0.97326	null	0.936197	null	0.895181	null	0.871133	null
GSE157657::response_group	cls	0.847108	null	0.857705	null	0.84723	null	0.835899	null	0.875277	null
GSE163882::pcr	cls	0.702728	null	0.735235	null	0.724752	null	0.702257	null	0.730291	null
GSE175718::rejection_type	cls	0.574639	null	0.577602	null	0.605117	null	0.555556	null	0.615195	null
GSE176178::bcg_response	cls	0.9075	null	0.881667	null	0.830833	null	0.865833	null	0.791667	null
GSE183250::metastasis	cls	0.8	null	0.85	null	0.8	null	0.85	null	0.85	null
GSE185263::mortality	cls	0.584262	null	0.574509	null	0.612537	null	0.610451	null	0.584102	null
GSE193677::historemiss	cls	0.710492	null	0.804054	null	0.762263	null	0.717265	null	0.681548	null
GSE198520::eular_response	cls	0.689418	null	0.694735	null	0.667725	null	0.71463	null	0.668995	null
GSE206505::vo2max_response	cls	0.844444	null	0.844444	null	0.844444	null	0.644444	null	0.822222	null
GSE213346::disease_course	cls	0.816667	null	0.833333	null	0.866667	null	0.825	null	0.833333	null
GSE234297::als_diagnosis	cls	0.68731	null	0.645322	null	0.656374	null	0.674591	null	0.646316	null
GSE234587::parasitemia	cls	0.672222	null	0.738889	null	0.472222	null	0.619444	null	0.572222	null
GSE243375::pcr	cls	0.656278	null	0.538576	null	0.632208	null	0.569924	null	0.634361	null
GSE244807::death	cls	0.681841	null	0.667689	null	0.675967	null	0.66752	null	0.688726	null
GSE278476::vaccine_response	cls	0.822222	null	0.822222	null	0.860317	null	0.764286	null	0.873016	null
GSE279480::cmv_status	cls	0.823922	null	0.823957	null	0.858557	null	0.782798	null	0.864792	null
GSE289743::pcr	cls	0.633333	null	0.933333	null	0.666667	null	0.933333	null	0.766667	null
GSE294705::hfri	cls	0.609091	null	0.422727	null	0.508182	null	0.428182	null	0.713636	null
GSE316750::bcr	cls	0.728333	null	0.578333	null	0.649167	null	0.646667	null	0.6575	null
GSE51799::carrier_status	cls	0.66895	null	0.649944	null	0.678056	null	0.676928	null	0.641019	null
GSE54460::bcr	cls	0.654949	null	0.650505	null	0.679596	null	0.663838	null	0.690707	null
GSE68086::cancer_type	cls	0.905458	null	0.90792	null	0.888851	null	0.86252	null	0.87605	null
GSE93624::complicated_progression	cls	0.693538	null	0.722163	null	0.711931	null	0.689123	null	0.769993	null
GTEx::hardy_scale	cls	0.918917	null	0.97635	null	0.930385	null	0.878759	null	0.802981	null
TARGET::inss_stage	cls	0.92358	null	0.924975	null	0.944193	null	0.927816	null	0.938778	null
TARGET::relapse	cls	0.763609	null	0.790014	null	0.751639	null	null	null	0.726876	null
TCGA::ajcc_stage_4	cls	0.735757	null	0.769796	null	0.699592	null	0.705505	null	0.707314	null
BeatAML::auc_Bortezomib	reg	null	0.131362	0.158286	0.198167	0.159671	0.16261	null	0.16344	0.232455	0.249036
BeatAML::auc_Dasatinib	reg	null	0.479518	0.507648	0.512765	0.528349	0.537826	null	0.469879	0.5277	0.530096
BeatAML::auc_Doramapimod	reg	null	0.267134	0.37195	0.390512	0.324173	0.328118	null	0.343711	0.38777	0.404111
BeatAML::auc_Gilteritinib	reg	null	0.265828	0.317467	0.296908	0.359594	0.370287	null	0.223817	0.322369	0.315049
BeatAML::auc_Midostaurin	reg	null	0.346737	0.398095	0.396905	0.383897	0.410943	null	0.262618	0.404013	0.398722
BeatAML::auc_Panobinostat	reg	null	0.509303	0.53333	0.516005	0.516008	0.495161	null	0.452121	0.500766	0.509459
BeatAML::auc_Quizartinib	reg	null	0.443743	0.500164	0.476253	0.533233	0.506152	null	0.479779	0.515278	0.488242
BeatAML::auc_Sorafenib	reg	null	0.522984	0.599084	0.546214	0.610038	0.580683	null	0.461739	0.626343	0.577369
BeatAML::auc_Trametinib	reg	null	0.61908	0.629992	0.637824	0.597013	0.606548	null	0.512517	0.612471	0.619144
BeatAML::auc_Venetoclax	reg	null	0.740353	0.764634	0.748179	0.767179	0.763704	null	0.659084	0.777904	0.770679
CMP::ic50_Bortezomib	reg	null	0.499342	0.459732	0.494002	0.355485	0.464659	null	0.422784	0.457403	0.509059
CMP::ic50_Doxorubicin	reg	null	0.519521	0.537239	0.541539	0.461418	0.526705	null	0.465148	0.54079	0.548523
CMP::ic50_Erlotinib	reg	null	0.53694	0.512928	0.530882	0.381418	0.493408	null	0.426262	0.480169	0.502986
CMP::ic50_Gemcitabine	reg	null	0.588099	0.61212	0.621056	0.494401	0.590204	null	0.530556	0.580596	0.595904
CMP::ic50_Olaparib	reg	null	0.658035	0.678323	0.664409	0.531285	0.623092	null	0.570243	0.649717	0.633755
CMP::ic50_Oxaliplatin	reg	null	0.657237	0.673717	0.663464	0.5032	0.625167	null	0.602623	0.660025	0.652236
CMP::ic50_Paclitaxel	reg	null	0.482456	0.476661	0.48493	0.301218	0.426614	null	0.437021	0.461243	0.480788
CMP::ic50_Rapamycin	reg	null	0.641534	0.650136	0.643906	0.50137	0.57022	null	0.542063	0.589155	0.594533
CMP::ic50_Ruxolitinib	reg	null	0.436509	0.46516	0.429612	0.419272	0.388863	null	0.369517	0.445983	0.418223
CMP::ic50_Vorinostat	reg	null	0.661671	0.634746	0.682652	0.585577	0.639312	null	0.632171	0.620543	0.677264
GSE102094::pfi	reg	0.342092	0.312857	0.21665	0.190476	0.413867	0.408095	0.134466	0.164762	0.404382	0.328095
GSE115525::prednisolone_lc50	reg	null	0.413161	0.147668	0.43096	0.150069	0.357974	null	0.301473	0.313225	0.44925
GSE157103::hospital_free_days	reg	null	0.542718	0.556025	0.579045	0.450445	0.494284	null	0.43615	0.514411	0.567688
GSE81538::ki67	reg	null	0.800833	0.808956	0.815509	0.80305	0.789248	null	0.7788	0.790752	0.822487
GTEx::telomere_length	reg	null	0.567854	0.662232	0.59074	0.626582	0.511815	null	0.546548	0.622427	0.525301

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Bulk RNA-seq benchmark for foundation-model evaluation: 58 tasks (25 regression, 33 classification) across 38 datasets spanning clinical outcome prediction, diagnostics, and biological/pharmacological phenotyping.

Format

One Parquet file per dataset in data/. Each row is a sample; columns are:

sample_id — sample accession (GSM / cell line / donor id).
gene columns — the native expression matrix X (gene-symbol headers).
y__<task> — one label column per task on that dataset. NaN = sample not in that task's cohort. Reproduce a task by selecting [gene cols] for rows where y__<task> is not NaN.
group__<task> — grouping key for tasks that use grouped cross-validation (donor / patient).

The number of usable rows in a task is the non-NaN count of its y__ column**, which is often smaller than the full cohort size

All expression is on a log1p scale and non-negative: mostly log1p-CPM; GTEx is log1p; CMP and GSE81538 have been converted to log1p from their native linear-TPM / log2 scales.

Loading

import pandas as pd
df = pd.read_parquet("data/CMP.parquet")
y = df["y__ic50_Bortezomib"]
mask = y.notna()
X = df.loc[mask, [c for c in df.columns if not c.startswith(("y__", "group__", "sample_id"))]]
y = y[mask]

Cross-validation splits

Most datasets have one sample per subject and can be split randomly. The following datasets contain multiple samples per patient/donor and must be split by their group__<task> column (GroupKFold / StratifiedGroupKFold), never randomly, or the same subject leaks across folds:

BeatAML, CPTAC, GSE107995, GSE156902, GSE157657, GSE193677, GSE243375, GSE278476, GSE279480, GTEx, TARGET, TCGA.

from sklearn.model_selection import GroupKFold
groups = df.loc[mask, "group__ajcc_stage_4"]          # e.g. TCGA patient barcode
for tr, te in GroupKFold(5).split(X, y, groups):
    ...

Task catalogue

Type A — Clinical & biological outcome prediction

Treatment response, recurrence, survival, drug sensitivity, staging, and molecular phenotypes.

Small (n ≤ 100)

Dataset	Task(s) — n	Description
GSE107422	`systemic_recurrence` (80, cls)	Colorectal cancer resection; prospective systemic recurrence
GSE115525	`prednisolone_lc50` (73, reg)	B-lineage ALL primary cells; prednisolone LC50 drug sensitivity
GSE116324	`bortezomib_response` (44, cls)	Multiple myeloma; bortezomib-induction response (≥VGPR sustained ≥1y)
GSE120622	`relapse` (80, cls)	NSCLC surgical tumor; prospective relapse
GSE154261	`recurrence` (73, cls)	T1 non-muscle-invasive bladder cancer post-reTUR; 24-month recurrence
GSE156699	`chemo_response` (88, cls)	High-grade serous ovarian cancer; platinum chemo response (PFS ≥6mo)
GSE157103	`hospital_free_days` (66, reg)	COVID-19 ICU leukocytes; hospital-free days (45-day window)
GSE157657	`response_group` (59, cls)	Blood TB; ATT treatment-response trajectory (4-class)
GSE176178	`bcg_response` (40, cls)	Stage-1 bladder cancer; durable vs non-durable BCG response
GSE198520	`eular_response` (46, cls)	RA synovium; EULAR response to anti-TNF (3-class)
GSE206505	`vo2max_response` (30, cls)	PBMC; VO2max responder vs non-responder (endurance RCT)
GSE213346	`disease_course` (40, cls)	Oral lichen planus biopsy; recalcitrant vs stable at 1 year
GSE278476	`vaccine_response` (46, cls)	PBMC MUC1-vaccine prevention trial; immune responder
GSE294705	`hfri` (60, cls)	NMIBC post-BCG; high-frequency recurrence index
GSE316750	`bcr` (41, cls)	Radical-prostatectomy tumor; biochemical recurrence
GSE54460	`bcr` (99, cls)	FFPE radical-prostatectomy; biochemical recurrence

Medium (n > 100)

Dataset	Task(s) — n	Description
BeatAML	`auc_Bortezomib` (461, reg), `auc_Dasatinib` (491, reg), `auc_Doramapimod` (452, reg), `auc_Gilteritinib` (277, reg), `auc_Midostaurin` (461, reg), `auc_Panobinostat` (285, reg), `auc_Quizartinib` (459, reg), `auc_Sorafenib` (492, reg), `auc_Trametinib` (482, reg), `auc_Venetoclax` (365, reg)	AML patient-derived ex-vivo drug sensitivity (AUC) for 10 targeted agents
CMP	`ic50_Bortezomib` (1195, reg), `ic50_Doxorubicin` (1179, reg), `ic50_Erlotinib` (1202, reg), `ic50_Gemcitabine` (1187, reg), `ic50_Olaparib` (1194, reg), `ic50_Oxaliplatin` (1179, reg), `ic50_Paclitaxel` (1196, reg), `ic50_Rapamycin` (1199, reg), `ic50_Ruxolitinib` (1177, reg), `ic50_Vorinostat` (1193, reg)	Pan-cancer cell lines (Cell Model Passports expression); IC50 drug sensitivity for 10 drugs with diverse MoAs
GSE109142	`week4_remission` (198, cls)	Pediatric ulcerative colitis (PROTECT); week-4 clinical remission
GSE112927	`graft_loss` (235, cls)	Pre-transplant blood; death-censored kidney allograft loss
GSE163882	`pcr` (217, cls)	Breast cancer pre-treatment biopsy; pCR to neoadjuvant chemotherapy
GSE185263	`mortality` (345, cls)	Early-sepsis whole blood; in-hospital mortality
GSE193677	`historemiss` (1301, cls)	Adult IBD biopsy (MSCCR); histological remission
GSE243375	`pcr` (242, cls)	HER2+ breast neoadjuvant T-DM1+pertuzumab; pathological complete response
GSE51799	`cag_repeats` (124, reg)	Huntington's disease blood; CAG-repeat length (severity) + mutation-carrier status
GSE93624	`complicated_progression` (245, cls)	Pediatric Crohn's (RISK) ileal biopsy; 3-year complicated (B2/B3) progression
GSE175718	`rejection_type` (234, cls)	Blood; kidney-transplant rejection type — ABMR / TCMR / polyoma / control (4-class)
GSE107995	`tb_group` (414, cls)	Whole-blood tuberculosis, future state — Active-TB / LTBI / Control / progressor (4-class)

Type B — Diagnostic prediction

Disease / infection presence or class predicted from the transcriptome (often a surrogate tissue such as blood or platelets).

Dataset	Task(s) — n	Description
CPTAC	`ajcc_stage_4` (2464, cls)	AJCC pathological stage (I–IV) from tumor RNA-seq
GSE156902	`group` (642, cls)	Tumor-educated platelets; GBM / brain-metastasis / MS / control (4-class)
GSE234297	`als_diagnosis` (144, cls)	Whole blood; ALS vs healthy diagnosis
GSE234587	`parasitemia` (33, cls)	Whole blood (Ugandan pregnant women); malaria parasitemia +/−
GSE279480	`cmv_status` (255, cls)	Whole blood; CMV serostatus positive / negative
GSE51799	`carrier_status` (124, cls)	Huntington's disease blood; CAG-repeat length (severity) + mutation-carrier status
GSE68086	`cancer_type` (246, cls)	Tumor-educated platelets; 7-class pan-cancer type vs healthy
GTEx	`telomere_length` (3940, reg)	Telomere length
TARGET	`inss_stage` (158, cls)	Pediatric neuroblastoma; INSS stage from tumor RNA-seq
TCGA	`ajcc_stage_4` (5206, cls)	Pan-cancer (33 types); AJCC pathological stage from tumor RNA-seq
GSE81538	`ki67` (405, reg)	SCAN-B breast cancer; Ki67 proliferation index (continuous)

Downloads last month: 128

Number of rows:

Total file size:

4.91 GB