OncoTraj
Public benchmark of 813 EGFR-mutant non-small-cell lung cancer patients on first-line osimertinib, harmonized from MSK-CHORD, AACR Project GENIE BPC and the FLAURA molecular-resistance supplement, with three locked tasks for longitudinal resistance prediction and an audited no-leakage split.
Composite
55.0
Experimental validation
N/A — retrospective real-world clinical-genomic benchmark
Stages
Clinical Development (cross-phase)Phase II
Modalities
small molecule
Task types
classificationregression
Size
patients: 813
tasks: 3
splits: {'train': 0, 'val': 0, 'test': 0}
note: 813 patients (MSK-CHORD 672, GENIE BPC 34, FLAURA supplement 107); patient-level train/val/test splits with audited no-leakage guarantee; 6 reference baselines
tasks: 3
splits: {'train': 0, 'val': 0, 'test': 0}
note: 813 patients (MSK-CHORD 672, GENIE BPC 34, FLAURA supplement 107); patient-level train/val/test splits with audited no-leakage guarantee; 6 reference baselines
License
Other — derived from MSK-CHORD / AACR GENIE BPC (controlled-access upstream)
First release
2026-06-09
Last updated
2026-06-09
Official site
Leaderboard
→ leaderboard
Dataset
Code / GitHub
→ repository
HuggingFace
→ HF
Paper
OncoTraj: a public benchmark for longitudinal resistance prediction in EGFR-mutant non-small-cell lung cancer on osimertinib · Abhijoy Sarkar, Aarchi Singh Thakur · 2026 · paper · doi:N/A — arXiv preprint 2606.11144 · 0 citations
Flags
none
Experts
Groups
Hosted by
—
Related benchmarks
Rubric (7-criterion)
rigor
4
coverage
3
maintenance
2
adoption
1
quality
4
accessibility
4
industry_relevance
3
Notes
Rare clinical-stage benchmark grounded in real-world clinical-genomic data with an explicitly audited no-leakage split (rigor 4, quality 4) and an open evaluation harness plus six reference baselines (accessibility 4). Authors are honest that v1 single-snapshot tissue NGS features clear no task above chance, correctly localizing the ceiling to input modality (needs serial ctDNA) rather than model class; that transparency is exactly the honesty the rubric rewards. Coverage limited to one drug/indication; adoption 0 (June 2026 preprint); upstream MSK-CHORD/GENIE access is controlled so reproducibility depends on credentialed data.