Structure & co-folding (7)
| # | Benchmark | Type | Evidence | Status | Models | Ranking |
|---|---|---|---|---|---|---|
| 01 | CASP Biennial blind test of protein structure prediction; remaining gains lie in complexes, RNA, ligands and ensembles. | A | L4 | Active | 15 | No ranking |
| 05 | PoseBusters Protein-ligand pose benchmark: an RMSD threshold plus 18 physical and chemical validity checks. | C+B | L3 | Active | 8 | Not comparable |
| 15 | CAMEO Weekly rolling blind test of structure-prediction servers on PDB pre-release sequences. | A | L4 | Active | 4 | Third-party |
| 22 | CAPRI The oldest prospective blind test for protein complexes; its classes and DockQ are the common protocol | A | L4 | Round in progress | 2 | No ranking |
| 25 | RNA-Puzzles Community blind test of RNA 3D structure; accuracy tracks template availability more than model type. | A+B | L2–L4 | Low-frequency | 3 | Board |
| 30 | PLINDER / PINDER / Runs N' Poses Docking and co-folding infrastructure that makes training-set similarity a first-class split criterion. | C+B | L3 | Active | 5 | Board |
| 31 | FoldBench All-atom co-folding evaluation on low-homology PDB assemblies; the antibody-antigen task discriminates most. | B | L3 | Low-frequency | 12 | Board |
Small molecules (10)
| # | Benchmark | Type | Evidence | Status | Models | Ranking |
|---|---|---|---|---|---|---|
| 03 | MoleculeNet (+OGB) Historical default suite for molecular representation learning: 17 public datasets, for literature alignment, not drug-discovery claims. | B | L2–L3 | Frozen | 2 | Not comparable |
| 04 | TDC (ADMET + DTI) De facto ADMET leaderboard; useful for method screening, not a basis for development decisions. | B | L2–L3 | Active | 4 | Board |
| 06 | CASF-2016 Decade-old default test of protein-ligand scoring functions across four powers; frozen and heavily overlapping with training data. | B | L2 | Frozen | 0 | Not comparable |
| 07 | USPTO family (+PaRoutes) Default data for retrosynthesis and reaction prediction; measures reproducing patent records, not lab feasibility. | B | L2–L3 | Frozen | 5 | Not comparable |
| 12 | DUD-E / LIT-PCBA Two generations of virtual-screening enrichment sets: constructed decoys (DUD-E) and PubChem inactives (LIT-PCBA); both have known bias or leakage. | B | L2 | Low-frequency | 2 | Third-party |
| 19 | PMO (Practical Molecular Optimization) Compares sample efficiency of molecular optimization within 10,000 oracle calls | D | L2 | Low-frequency | 2 | Self-reported |
| 20 | CACHE Teams submit purchasable compounds and the organizers measure binding: a blind test with real assays | A | L5 | Active | 0 | No ranking |
| 27 | Polaris / ASAP / OpenADMET Industry-led benchmark hub, guidelines and blind challenges; its flagship is an antiviral challenge on unpublished project data. | A+C | L4 | Low-frequency | 0 | No ranking |
| 33 | PoseX Compares 23 physics, AI-docking and co-folding methods under one protocol, separating self-docking from cross-docking. | B | L3 | Active | 6 | Third-party |
| 44 | FS-Mol With only 16-256 labels on a brand-new assay, predict the rest; it fills the few-shot adaptation gap. | B | L3 | Frozen | 0 | Not comparable |
Proteins (5)
| # | Benchmark | Type | Evidence | Status | Models | Ranking |
|---|---|---|---|---|---|---|
| 02 | ProteinGym Unified suite of 200+ public deep mutational scans for protein variant-effect and fitness ranking. | B | L2–L3 | Low-frequency | 10 | Board |
| 17 | CAFA Time-delayed blind test of protein function (GO) prediction against experimental annotations added after submission. | A | L4–L5 | Active | 0 | No ranking |
| 26 | FLIP / FLIP2 Supervised protein fitness benchmark using deliberately shifted splits to test low-resource learning and extrapolation. | B | L2–L3 | Low-frequency | 2 | Not comparable |
| 28 | Adaptyv competitions / BenchBB Multi-round prospective protein design evaluation with public data; run by a commercial service provider, not a board to optimise against. | A+B | L5 | Active | 4 | No ranking |
| 40 | MotifBench Thirty fixed motif-scaffolding problems judged by one pipeline, with a submission leaderboard; ground truth stays computational. | D | L2 | Active | 3 | Board |
Antibodies (4)
| # | Benchmark | Type | Evidence | Status | Models | Ranking |
|---|---|---|---|---|---|---|
| 34 | AbBiBench Zero-shot scoring correlations evaluated on antibody-antigen complexes, filling a gap in antibody function evaluation. | B | L2–L3 | Active | 6 | Board |
| 35 | FLAb / FLAb2 Aggregated retrospective antibody developability benchmark with a negative headline result. | B | L2–L3 | Active | 1 | Third-party |
| 36 | AIntibody Antibody-side prospective blind test: submitted designs get uniform affinity and developability measurement. | A | L5 | Low-frequency | 1 | Board |
| 43 | Ginkgo GDPa1 Uniform-assay developability data on 246 clinical antibodies plus a one-off blind test on 80 private held-out antibodies. | A+B | L4 | Low-frequency | 0 | Board |
Genomics & RNA (8)
| # | Benchmark | Type | Evidence | Status | Models | Ranking |
|---|---|---|---|---|---|---|
| 08 | Basenji/Enformer test set The de facto protocol for sequence-to-function models: track prediction on fixed held-out regions. | B | L2–L3 | Active | 3 | Self-reported |
| 11 | ClinVar hold-out protocol Separating pathogenic from benign on held-out clinical labels, with each paper choosing its own snapshot and subset. | B | L2–L3 | Active | 5 | Not comparable |
| 13 | CAGI Long-running CASP-style blind test of variant interpretation, scored on data unpublished at submission. | A | L4–L5 | Active | 1 | Third-party |
| 16 | GUE + Nucleotide Transformer benchmark Short-sequence regulatory-element classification suites for DNA language models; a smoke test, not proof of quality. | B | L2–L3 | Low-frequency | 0 | No ranking |
| 21 | DREAM5 / BEELINE / CausalBench Three generations of GRN inference evaluation whose scores mainly reflect gold-standard quality | A+B+C | L2–L5 | Low-frequency | 0 | Not comparable |
| 37 | TraitGym Benchmark classifying causal non-coding variants against matched controls. | B | L3 | Low-frequency | 3 | Board |
| 38 | BEND Evaluates DNA language models' frozen representations on human-genome functional tasks. | B | L2 | Active | 0 | Board |
| 39 | RNAGym Zero-shot evaluation framework for RNA fitness and structure. | B | L2 | Active | 1 | Board |
Single-cell & perturbation (2)
| # | Benchmark | Type | Evidence | Status | Models | Ranking |
|---|---|---|---|---|---|---|
| 09 | scIB (+Open Problems) Default yardstick for single-cell integration, scoring batch removal against biological conservation; continued by Open Problems. | B+C | L2 | Active | 1 | Board |
| 24 | Virtual Cell Challenge and perturbation benchmarks The only challenge generating Perturb-seq data for evaluation, yet linear and mean baselines remain hard to beat | A+B+C | L2–L4 | Active | 9 | Board |
Molecule generation (2)
| # | Benchmark | Type | Evidence | Status | Models | Ranking |
|---|---|---|---|---|---|---|
| 10 | GuacaMol (+MOSES) Entry-level molecular generation benchmark; computed oracles only, unmaintained, optimization side superseded by PMO. | D | L2 | Frozen | 0 | No ranking |
| 32 | CrossDocked2020 The de facto default set for pocket-conditioned molecule generation, scored by docking software. | D | L2 | Frozen | 1 | Not comparable |
Text, LLMs & agents (6)
| # | Benchmark | Type | Evidence | Status | Models | Ranking |
|---|---|---|---|---|---|---|
| 14 | PubMedQA 500-question yes/no/maybe test on PubMed abstracts; top scores now sit within sampling noise. | B+E | L1 | Frozen | 2 | Board |
| 18 | LAB-Bench / LABBench2 Biology research-skill benchmark for LLMs and agents; the multiple-choice original is partly saturated and superseded by LABBench2. | E | L1–L2 | Active | 3 | Board |
| 23 | BioASQ The only yearly biomedical retrieval and QA challenge with new expert-written questions and expert grading | A+B | L1–L4 | Active | 0 | Board |
| 29 | BixBench Agents write and run code on real data in a container to do bioinformatics analyses, graded on their conclusions. | E | L2 | Active | 3 | Board |
| 41 | BioML-bench Agents build and run biomedical ML pipelines end to end, scored against human public-leaderboard results; tested agents average below the human baseline. | E | L2 | Low-frequency | 1 | Self-reported |
| 42 | BioMysteryBench Questions on real sequencing data with metadata stripped, graded on the final answer only; authored by a model vendor, with only five public questions. | E | L2–L4 | Active | 1 | Third-party |