{ "version": "1.0.0", "generated": "2026-05-12", "count": 73, "benchmarks": [ { "id": "open-targets", "name": "Open Targets Platform", "stages": [ "disease-modeling", "target-id" ], "modalities": [ "cross-modality" ], "task_types": [ "target-disease-association", "prioritization" ], "description": "Integrates genetic, genomic, pharmacological evidence to score targets for 20k+ diseases.", "size": { "targets": 63000, "diseases": 28000, "evidence": 18000000 }, "primary_paper": { "title": "The Open Targets Platform: supporting systematic drug-target identification and prioritisation", "authors": [ "Ochoa D", "Hercules A", "Carmona M", "et al." ], "year": 2021, "doi": "10.1093/nar/gkaa1027", "citations": 1100 }, "official_url": "https://platform.opentargets.org/", "github_url": "https://github.com/opentargets", "leaderboard_url": "N/A", "license": "CC0 / Apache-2.0", "first_release": "2016-12", "last_updated": "2025-06", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 5, "adoption": 5, "quality": 5, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Industry gold standard for target prioritization. Quarterly versioned releases.", "related_benchmarks": [ "depmap", "disgenet", "primekg" ], "expert_ids": [ "ian-dunham", "ellen-mcdonagh" ], "group_ids": [ "opentargets", "embl-ebi", "gsk", "sanger" ], "hosted_by": [ "elixir" ], "composite_score": 100.0 }, { "id": "depmap", "name": "DepMap (Cancer Dependency Map)", "stages": [ "target-id", "disease-modeling" ], "modalities": [ "cross-modality" ], "task_types": [ "essentiality", "biomarker" ], "description": "Genome-scale CRISPR/RNAi essentiality across 1150 cancer cell lines + omics.", "size": { "cell_lines": 1150, "genes": 18000 }, "primary_paper": { "title": "Defining a Cancer Dependency Map", "authors": [ "Tsherniak A", "Vazquez F", "Montgomery PG", "et al." ], "year": 2017, "doi": "10.1016/j.cell.2017.06.010", "citations": 2600 }, "official_url": "https://depmap.org/portal/", "github_url": "https://github.com/broadinstitute/depmap", "leaderboard_url": "https://depmap.org/portal/prediction/", "license": "CC-BY 4.0", "first_release": "2017-07", "last_updated": "2025-06", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 5, "adoption": 5, "quality": 5, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Quarterly release cadence.", "related_benchmarks": [ "open-targets", "lincs-l1000" ], "expert_ids": [ "aviad-tsherniak", "william-hahn", "todd-golub" ], "group_ids": [ "broad-depmap" ], "hosted_by": [], "composite_score": 100.0 }, { "id": "tdc-admet", "name": "TDC ADMET Group", "stages": [ "lead-id-admet" ], "modalities": [ "small-molecule" ], "task_types": [ "regression", "classification" ], "description": "22-task ADMET benchmark suite with scaffold splits \u2014 core TDC leaderboard.", "size": { "tasks": 22, "molecules": 130000 }, "primary_paper": { "title": "Therapeutics Data Commons: Machine Learning Datasets and Tasks for Drug Discovery and Development", "authors": [ "Huang K", "Fu T", "Gao W", "et al." ], "year": 2021, "doi": "10.48550/arXiv.2102.09548", "citations": 620 }, "official_url": "https://tdcommons.ai/benchmark/admet_group/overview/", "github_url": "https://github.com/mims-harvard/TDC", "leaderboard_url": "https://tdcommons.ai/benchmark/admet_group/overview/", "license": "MIT", "first_release": "2021-02", "last_updated": "2025-04", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 5, "adoption": 5, "quality": 5, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Most-adopted ADMET benchmark. 100+ leaderboard submissions.", "related_benchmarks": [ "moleculenet", "admet-ai", "polaris-admet", "molecule-ace" ], "expert_ids": [ "kexin-huang", "marinka-zitnik", "tianfan-fu" ], "group_ids": [ "zitnik-lab", "tdc" ], "hosted_by": [ "tdc", "insilico-scienceaibench", "insilico-ddb" ], "composite_score": 100.0 }, { "id": "sabdab", "name": "SAbDab", "stages": [ "hit-id", "lead-id-admet", "developmental-candidate" ], "modalities": [ "biologic-mab" ], "task_types": [ "antibody-structure", "affinity" ], "description": "Structural antibody database \u2014 curated PDB antibody structures with annotation.", "size": { "structures": 9500, "complexes": 2600 }, "primary_paper": { "title": "SAbDab: the structural antibody database", "authors": [ "Dunbar J", "Krawczyk K", "Leem J", "et al." ], "year": 2014, "doi": "10.1093/nar/gkt1043", "citations": 630 }, "official_url": "https://opig.stats.ox.ac.uk/webapps/newsabdab/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "CC-BY 4.0", "first_release": "2013", "last_updated": "2025-04", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 5, "adoption": 5, "quality": 5, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Canonical antibody structure resource. Weekly updates.", "related_benchmarks": [ "oas", "cov-abdab", "iglm-bench" ], "expert_ids": [ "charlotte-deane", "james-dunbar" ], "group_ids": [ "oxford-opig" ], "hosted_by": [], "composite_score": 100.0 }, { "id": "chembl", "name": "ChEMBL", "stages": [ "hit-id", "lead-id-admet" ], "modalities": [ "small-molecule", "protein-general" ], "task_types": [ "bioactivity", "data-resource" ], "description": "Manually curated bioactive molecule DB; backbone for most ML chemistry benchmarks.", "size": { "compounds": 2400000, "activities": 20700000, "targets": 15398 }, "primary_paper": { "title": "The ChEMBL Database in 2023", "authors": [ "Zdrazil B", "Felix E", "Hunter F", "et al." ], "year": 2024, "doi": "10.1093/nar/gkad1004", "citations": 800 }, "official_url": "https://www.ebi.ac.uk/chembl/", "github_url": "https://github.com/chembl", "leaderboard_url": "N/A", "license": "CC-BY-SA 3.0", "first_release": "2009", "last_updated": "2025-05", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 5, "adoption": 5, "quality": 4, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Underlies ~80% of public bioactivity ML benchmarks.", "related_benchmarks": [ "tdc-admet", "moleculenet", "pubchem-bioassay" ], "expert_ids": [ "andrew-leach", "barbara-zdrazil" ], "group_ids": [ "embl-ebi" ], "hosted_by": [ "elixir" ], "composite_score": 97.5 }, { "id": "oas", "name": "Observed Antibody Space (OAS)", "stages": [ "hit-id", "lead-id-admet" ], "modalities": [ "biologic-mab" ], "task_types": [ "antibody-sequence" ], "description": "Repository of >1B antibody BCR sequences from public repertoires \u2014 core for antibody LM pretraining.", "size": { "sequences": 2400000000, "repertoires": 15000 }, "primary_paper": { "title": "Observed Antibody Space: A diverse database of cleaned, annotated, and translated unpaired and paired antibody sequences", "authors": [ "Olsen TH", "Boyles F", "Deane CM" ], "year": 2022, "doi": "10.1002/pro.4205", "citations": 310 }, "official_url": "https://opig.stats.ox.ac.uk/webapps/oas/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "CC-BY 4.0", "first_release": "2018", "last_updated": "2025-04", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 5, "adoption": 5, "quality": 4, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Underlies AbLang, IgLM, AntiBERTa \u2014 industry-adopted.", "related_benchmarks": [ "sabdab", "cov-abdab", "iglm-bench" ], "expert_ids": [ "charlotte-deane", "tobias-olsen" ], "group_ids": [ "oxford-opig" ], "hosted_by": [], "composite_score": 97.5 }, { "id": "proteingym", "name": "ProteinGym", "stages": [ "target-id", "lead-id-admet", "ind-enabling" ], "modalities": [ "protein-general" ], "task_types": [ "variant-effect" ], "description": "217 DMS substitution assays + indel + clinical variants \u2014 de facto standard for VEPs.", "size": { "dms_assays": 217, "mutations": 2700000, "clinical_variants": 2525 }, "primary_paper": { "title": "ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design", "authors": [ "Notin P", "Kollasch A", "Ritter D", "et al." ], "year": 2023, "doi": "10.48550/arXiv.2305.06259", "citations": 320 }, "official_url": "https://proteingym.org/", "github_url": "https://github.com/OATML-Markslab/ProteinGym", "leaderboard_url": "https://proteingym.org/benchmarks", "license": "MIT", "first_release": "2022", "last_updated": "2025-03", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 5, "adoption": 5, "quality": 5, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Field standard. Clinical track enables fair ESM/EVE/AlphaMissense comparison.", "related_benchmarks": [ "flip", "cafa-benchmark" ], "expert_ids": [ "debora-marks", "pascal-notin", "yarin-gal" ], "group_ids": [ "marks-lab", "oatml" ], "hosted_by": [ "proteingym" ], "composite_score": 97.5 }, { "id": "posebusters", "name": "PoseBusters", "stages": [ "hit-id" ], "modalities": [ "small-molecule", "protein-general" ], "task_types": [ "pose-validation" ], "description": "Physics-aware eval of docking/co-folding poses \u2014 19 checks catching chemically impossible outputs.", "size": { "complexes": 428, "checks_per_pose": 19 }, "primary_paper": { "title": "PoseBusters: AI-based docking methods fail to generate physically valid poses or generalise to novel sequences", "authors": [ "Buttenschoen M", "Morris GM", "Deane CM" ], "year": 2024, "doi": "10.1039/D3SC04185A", "citations": 360 }, "official_url": "https://posebusters.readthedocs.io/", "github_url": "https://github.com/maabuu/posebusters", "leaderboard_url": "https://github.com/maabuu/posebusters", "license": "BSD-3-Clause", "first_release": "2023-08", "last_updated": "2025-02", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 5, "adoption": 5, "quality": 5, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Exposed major failure modes in AlphaFold-Multimer/DiffDock/RFAA. Default pharma filter.", "related_benchmarks": [ "plinder", "pinder", "casf-2016" ], "expert_ids": [ "charlotte-deane", "martin-buttenschoen" ], "group_ids": [ "oxford-opig" ], "hosted_by": [ "posebusters-initiative" ], "composite_score": 97.0 }, { "id": "plinder", "name": "PLINDER", "stages": [ "hit-id" ], "modalities": [ "small-molecule", "protein-general" ], "task_types": [ "docking", "structure-based-benchmark" ], "description": "Leakage-aware protein-ligand interaction dataset with stratified splits.", "size": { "complexes": 1400000, "unique_systems": 460000 }, "primary_paper": { "title": "PLINDER: The protein-ligand interactions dataset and evaluation resource", "authors": [ "Durairaj J", "Adeshina Y", "Cao Z", "et al." ], "year": 2024, "doi": "10.48550/arXiv.2409.17475", "citations": 55 }, "official_url": "https://www.plinder.sh/", "github_url": "https://github.com/plinder-org/plinder", "leaderboard_url": "N/A", "license": "CC-BY 4.0", "first_release": "2024-09", "last_updated": "2025-03", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 5, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Replaces PDBbind as the modern leakage-controlled docking standard.", "related_benchmarks": [ "pdbbind", "pinder", "posebusters" ], "expert_ids": [ "torsten-schwede", "max-jaderberg" ], "group_ids": [ "biozentrum-basel", "isomorphic-labs", "vantai" ], "hosted_by": [ "plinder-initiative" ], "composite_score": 97.0 }, { "id": "stringdb", "name": "STRING", "stages": [ "target-id", "disease-modeling" ], "modalities": [ "protein-general" ], "task_types": [ "ppi", "network-inference" ], "description": "Protein-protein interaction & functional association network across 12k organisms.", "size": { "proteins": 67000000, "associations": 20000000000, "organisms": 12535 }, "primary_paper": { "title": "The STRING database in 2023", "authors": [ "Szklarczyk D", "Kirsch R", "Koutrouli M", "et al." ], "year": 2023, "doi": "10.1093/nar/gkac1000", "citations": 2800 }, "official_url": "https://string-db.org/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "CC-BY 4.0", "first_release": "2000", "last_updated": "2024-11", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 5, "adoption": 5, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Workhorse for network-based target ID. Distinguish functional vs physical edges.", "related_benchmarks": [ "open-targets", "primekg" ], "expert_ids": [ "christian-von-mering" ], "group_ids": [ "sib-swiss" ], "hosted_by": [ "elixir" ], "composite_score": 94.9 }, { "id": "casp15", "name": "CASP15", "stages": [ "hit-id", "target-id" ], "modalities": [ "protein-general" ], "task_types": [ "structure-prediction" ], "description": "CASP15 blind structure prediction including multimer and ligand-bound categories.", "size": { "targets": 127, "categories": 5 }, "primary_paper": { "title": "Critical assessment of methods of protein structure prediction (CASP) \u2014 Round XV", "authors": [ "Kryshtafovych A", "Schwede T", "Topf M", "et al." ], "year": 2023, "doi": "10.1002/prot.26617", "citations": 140 }, "official_url": "https://predictioncenter.org/casp15/", "github_url": "N/A", "leaderboard_url": "https://predictioncenter.org/casp15/results", "license": "Public", "first_release": "2022", "last_updated": "2023", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 3, "adoption": 5, "quality": 5, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Biennial. Introduced ligand prediction category.", "related_benchmarks": [ "casp16", "cameo-targets" ], "expert_ids": [ "andriy-kryshtafovych", "john-moult", "torsten-schwede" ], "group_ids": [ "prediction-center-ucd" ], "hosted_by": [ "casp" ], "composite_score": 94.9 }, { "id": "casp16", "name": "CASP16", "stages": [ "hit-id" ], "modalities": [ "protein-general", "small-molecule" ], "task_types": [ "structure-prediction", "ligand-pose" ], "description": "CASP16 (2024) \u2014 expanded multimer, RNA, and ligand prediction.", "size": { "targets": 140, "categories": 6 }, "primary_paper": { "title": "CASP16 preliminary overview", "authors": "CASP16 assessors", "year": 2024, "doi": "N/A \u2014 overview papers pending", "citations": 15 }, "official_url": "https://predictioncenter.org/casp16/", "github_url": "N/A", "leaderboard_url": "https://predictioncenter.org/casp16/results", "license": "Public", "first_release": "2024", "last_updated": "2024-12", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 4, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "First full multimer+ligand+RNA joint eval.", "related_benchmarks": [ "casp15" ], "expert_ids": [ "andriy-kryshtafovych", "john-moult" ], "group_ids": [ "prediction-center-ucd" ], "hosted_by": [ "casp" ], "composite_score": 94.4 }, { "id": "cameo-targets", "name": "CAMEO weekly targets", "stages": [ "hit-id" ], "modalities": [ "protein-general", "small-molecule" ], "task_types": [ "structure-prediction", "ligand-pose" ], "description": "Continuous weekly blind eval using pre-release PDB structures \u2014 3D, multimer, ligand pocket.", "size": { "targets_per_year": 1000, "categories": 4 }, "primary_paper": { "title": "CAMEO: continuous evaluation of computational biology methods", "authors": [ "Haas J", "Roth S", "Arnold K", "et al." ], "year": 2013, "doi": "10.1093/database/bat031", "citations": 220 }, "official_url": "https://www.cameo3d.org/", "github_url": "N/A", "leaderboard_url": "https://www.cameo3d.org/", "license": "CC-BY 4.0", "first_release": "2013", "last_updated": "2025-05", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 5, "adoption": 5, "quality": 5, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Weekly cadence complements biennial CASP.", "related_benchmarks": [ "casp15", "casp16" ], "expert_ids": [ "torsten-schwede", "jurgen-haas" ], "group_ids": [ "biozentrum-basel" ], "hosted_by": [ "cameo" ], "composite_score": 94.4 }, { "id": "ord-bench", "name": "ORD Reaction Benchmark", "stages": [ "developmental-candidate" ], "modalities": [ "small-molecule" ], "task_types": [ "reaction-prediction", "yield" ], "description": "Reaction benchmarks derived from the Open Reaction Database (yield prediction, condition recommendation).", "size": { "reactions": 2100000, "tasks": 4 }, "primary_paper": { "title": "The Open Reaction Database", "authors": [ "Kearnes SM", "Maser MR", "Wleklinski M", "et al." ], "year": 2021, "doi": "10.1021/jacs.1c09820", "citations": 210 }, "official_url": "https://open-reaction-database.org/", "github_url": "https://github.com/open-reaction-database", "leaderboard_url": "N/A", "license": "CC-BY-SA 4.0", "first_release": "2021-07", "last_updated": "2025-04", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 5, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Modern open reaction corpus; industry-scale.", "related_benchmarks": [ "uspto-retrosyn" ], "expert_ids": [ "steven-kearnes", "abigail-doyle", "connor-coley" ], "group_ids": [ "google-research", "princeton-chemistry", "coley-lab" ], "hosted_by": [ "open-reaction-database" ], "composite_score": 93.9 }, { "id": "openproblems-perturbation", "name": "Open Problems: Perturbation Prediction", "stages": [ "virtual-cell" ], "modalities": [ "cross-modality" ], "task_types": [ "perturbation-prediction" ], "description": "NeurIPS competition & continuing benchmark for single-cell perturbation response prediction under distribution shift.", "size": { "cells": 240000, "perturbations": 144 }, "primary_paper": { "title": "Predicting Cellular Responses to Novel Drug Perturbations at a Single-Cell Resolution", "authors": [ "Hetzel L", "Boehm S", "Kilbertus N", "et al." ], "year": 2022, "doi": "10.48550/arXiv.2204.13545", "citations": 180 }, "official_url": "https://openproblems.bio/", "github_url": "https://github.com/openproblems-bio/openproblems", "leaderboard_url": "https://openproblems.bio/results", "license": "MIT", "first_release": "2021-06", "last_updated": "2024-12", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 5, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 3 }, "flags": [], "notes": "Best-in-class rigor (Viash workflow, hidden test, NeurIPS track).", "related_benchmarks": [ "cz-virtual-cell-challenge", "scperturb", "lincs-l1000" ], "expert_ids": [ "fabian-theis", "daniel-burkhardt", "malte-luecken" ], "group_ids": [ "openproblems", "helmholtz-munich", "czi-science" ], "hosted_by": [ "openproblems", "czi-virtual-cell" ], "composite_score": 91.9 }, { "id": "primekg", "name": "PrimeKG", "stages": [ "disease-modeling", "target-id" ], "modalities": [ "cross-modality" ], "task_types": [ "knowledge-graph", "drug-repurposing" ], "description": "Precision medicine knowledge graph integrating 20 sources for 17k diseases.", "size": { "nodes": 129375, "edges": 8100498, "diseases": 17080 }, "primary_paper": { "title": "Building a knowledge graph to enable precision medicine", "authors": [ "Chandak P", "Huang K", "Zitnik M" ], "year": 2023, "doi": "10.1038/s41597-023-01960-3", "citations": 320 }, "official_url": "https://zitniklab.hms.harvard.edu/projects/PrimeKG/", "github_url": "https://github.com/mims-harvard/PrimeKG", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2023-03", "last_updated": "2025-02", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 4, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Modern, well-engineered KG; strong for GNN drug repurposing.", "related_benchmarks": [ "open-targets", "disgenet" ], "expert_ids": [ "marinka-zitnik", "payal-chandak", "kexin-huang" ], "group_ids": [ "zitnik-lab" ], "hosted_by": [ "tdc" ], "composite_score": 91.9 }, { "id": "faers-bench", "name": "FAERS (raw)", "stages": [ "post-market-rwe" ], "modalities": [ "small-molecule", "biologic-mab" ], "task_types": [ "pharmacovigilance" ], "description": "FDA Adverse Event Reporting System \u2014 19M+ reports used as substrate for signal-detection ML.", "size": { "reports": 19000000 }, "primary_paper": { "title": "An assessment of the U.S. FDA Adverse Event Reporting System (FAERS) and the impact of quality reporting", "authors": [ "Sakaeda T", "Tamon A", "Kadoyama K", "Okuno Y" ], "year": 2013, "doi": "10.3390/ijerph100300796", "citations": 780 }, "official_url": "https://open.fda.gov/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "Public domain", "first_release": "1969", "last_updated": "2025-Q2", "rubric": { "rigor": 4, "coverage": 5, "maintenance": 5, "adoption": 5, "quality": 3, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Known under-/over-reporting biases.", "related_benchmarks": [ "offsides-twosides", "sider" ], "expert_ids": [ "FDA CDER" ], "group_ids": [ "fda-cder" ], "hosted_by": [ "faers" ], "composite_score": 91.1 }, { "id": "longevity-bench-insilico", "name": "Longevity Benchmark (Insilico)", "stages": [ "disease-modeling", "target-id", "post-market-rwe" ], "modalities": [ "cross-modality" ], "task_types": [ "aging-prediction", "survival", "biomarker" ], "description": "19 aging/longevity benchmarks on ScienceAIBench/InsilicoBench/DDB \u2014 NHANES mortality, TCGA survival, methylation age, GTEx, Olink proteomic, longevity synergy.", "size": { "benchmarks": 19, "datasets": [ "NHANES", "TCGA", "Methylation", "GTEx", "Olink" ] }, "primary_paper": { "title": "Longevity Benchmark methodology", "authors": "Insilico AI Team", "year": 2025, "doi": "N/A \u2014 portal", "citations": 12 }, "official_url": "https://insilicobench.insilico.com/", "github_url": "N/A", "leaderboard_url": "https://insilicobench.insilico.com/", "license": "CC-BY", "first_release": "2025-02", "last_updated": "2026-04", "rubric": { "rigor": 4, "coverage": 5, "maintenance": 5, "adoption": 4, "quality": 4, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Unique, broad longevity/aging benchmark slice \u2014 nothing else in the field covers aging comparably. Leaderboard features frontier LLMs.", "related_benchmarks": [ "depmap", "mimic-benchmark" ], "expert_ids": [ "alex-zhavoronkov", "alex-aliper", "alex-zhebrak" ], "group_ids": [ "insilico-medicine" ], "hosted_by": [ "insilico-scienceaibench", "insilico-insilicobench", "insilico-ddb" ], "composite_score": 90.6 }, { "id": "lincs-l1000", "name": "LINCS L1000 / CMap", "stages": [ "virtual-cell", "disease-modeling", "target-id" ], "modalities": [ "small-molecule", "cross-modality" ], "task_types": [ "signature-match", "mechanism-of-action" ], "description": "1.3M transcriptional profiles across 978 landmark genes after genetic or chemical perturbations.", "size": { "profiles": 1319138, "compounds": 33000, "cell_lines": 82 }, "primary_paper": { "title": "A Next Generation Connectivity Map: L1000 Platform and the First 1,000,000 Profiles", "authors": [ "Subramanian A", "Narayan R", "Corsello SM", "et al." ], "year": 2017, "doi": "10.1016/j.cell.2017.10.049", "citations": 3700 }, "official_url": "https://clue.io/", "github_url": "https://github.com/cmap", "leaderboard_url": "N/A", "license": "CC-BY", "first_release": "2017-11", "last_updated": "2024-03", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 3, "adoption": 5, "quality": 4, "accessibility": 4, "industry_relevance": 5 }, "flags": [], "notes": "Foundational pharma resource for MoA work. Batch effects require careful handling.", "related_benchmarks": [ "scperturb", "depmap" ], "expert_ids": [ "aravind-subramanian", "todd-golub" ], "group_ids": [ "broad-cmap" ], "hosted_by": [], "composite_score": 89.9 }, { "id": "cansar", "name": "canSAR", "stages": [ "target-id", "hit-id" ], "modalities": [ "small-molecule", "protein-general" ], "task_types": [ "druggability", "target-annotation" ], "description": "Cancer translational knowledge base integrating pharmacology, bioactivity, structure, druggability.", "size": { "proteins": 25000, "compounds": 11000000, "bioactivity_datapoints": 106000000 }, "primary_paper": { "title": "canSAR: update to the cancer translational research and drug discovery knowledgebase", "authors": [ "Mitsopoulos C", "Di Micco P", "et al." ], "year": 2021, "doi": "10.1093/nar/gkaa1059", "citations": 210 }, "official_url": "https://cansar.ai/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "Academic free / commercial tier", "first_release": "2011", "last_updated": "2024-09", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 4, "adoption": 4, "quality": 5, "accessibility": 3, "industry_relevance": 5 }, "flags": [], "notes": "Deep oncology focus; widely-used druggability predictor.", "related_benchmarks": [ "open-targets", "depmap", "chembl" ], "expert_ids": [ "bissan-al-lazikani" ], "group_ids": [ "icr-london" ], "hosted_by": [], "composite_score": 89.4 }, { "id": "mimic-benchmark", "name": "MIMIC-IV Benchmark Tasks", "stages": [ "phase-3", "clinical-development", "post-market-rwe" ], "modalities": [ "cross-modality" ], "task_types": [ "clinical-outcome" ], "description": "Standardized ICU benchmarks on MIMIC-IV \u2014 mortality, LOS, sepsis, AKI, drug dosing.", "size": { "patients": 299712, "tasks": 14 }, "primary_paper": { "title": "MIMIC-IV, a freely accessible electronic health record dataset", "authors": [ "Johnson AEW", "Bulgarelli L", "Shen L", "et al." ], "year": 2023, "doi": "10.1038/s41597-022-01899-x", "citations": 620 }, "official_url": "https://physionet.org/content/mimiciv/", "github_url": "https://github.com/MIT-LCP/mimic-code", "leaderboard_url": "N/A", "license": "PhysioNet Credentialed", "first_release": "2020", "last_updated": "2025-02", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 5, "adoption": 5, "quality": 5, "accessibility": 3, "industry_relevance": 4 }, "flags": [], "notes": "Canonical clinical ML benchmark. Credentialed access limits casual use.", "related_benchmarks": [ "ctod" ], "expert_ids": [ "alistair-johnson", "leo-anthony-celi", "roger-mark" ], "group_ids": [ "mit-lcp" ], "hosted_by": [ "mimic" ], "composite_score": 89.4 }, { "id": "scperturb", "name": "scPerturb", "stages": [ "virtual-cell", "target-id" ], "modalities": [ "cross-modality" ], "task_types": [ "perturbation-prediction" ], "description": "Harmonized single-cell perturbation datasets (genetic + chemical) with standardized metadata.", "size": { "cells": 1700000, "datasets": 44 }, "primary_paper": { "title": "scPerturb: Harmonized single-cell perturbation data", "authors": [ "Peidli S", "Green TD", "Shen C", "Theis FJ", "et al." ], "year": 2024, "doi": "10.1038/s41592-023-02144-y", "citations": 280 }, "official_url": "http://projects.sanderlab.org/scperturb/", "github_url": "https://github.com/sanderlab/scPerturb", "leaderboard_url": "N/A \u2014 dataset resource", "license": "MIT", "first_release": "2023-05", "last_updated": "2024-11", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 4, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Canonical harmonized resource. Strong Perturb-seq coverage; weaker for chemical perturbations.", "related_benchmarks": [ "cz-virtual-cell-challenge", "openproblems-perturbation", "lincs-l1000" ], "expert_ids": [ "chris-sander", "fabian-theis" ], "group_ids": [ "sander-lab", "helmholtz-munich" ], "hosted_by": [], "composite_score": 88.9 }, { "id": "pinder", "name": "PINDER", "stages": [ "hit-id" ], "modalities": [ "protein-general" ], "task_types": [ "protein-protein-docking" ], "description": "Protein-protein docking benchmark with rigorous split design.", "size": { "dimers": 2319564, "systems": 267498 }, "primary_paper": { "title": "PINDER: The Protein Interaction Dataset and Evaluation Resource", "authors": [ "Kovtun D", "Akdel M", "Goncearenco A", "et al." ], "year": 2024, "doi": "10.1101/2024.07.17.603980", "citations": 45 }, "official_url": "https://www.pinder.sh/", "github_url": "https://github.com/pinder-org/pinder", "leaderboard_url": "N/A", "license": "CC-BY 4.0", "first_release": "2024-07", "last_updated": "2025-03", "rubric": { "rigor": 5, "coverage": 5, "maintenance": 4, "adoption": 3, "quality": 5, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Expected PPI docking standard.", "related_benchmarks": [ "plinder", "capri-benchmark" ], "expert_ids": [ "torsten-schwede" ], "group_ids": [ "vantai", "biozentrum-basel" ], "hosted_by": [ "plinder-initiative" ], "composite_score": 88.9 }, { "id": "pmo", "name": "Practical Molecular Optimization (PMO)", "stages": [ "lead-id-admet", "developmental-candidate" ], "modalities": [ "small-molecule" ], "task_types": [ "molecule-generation", "oracle" ], "description": "Budget-aware benchmark for goal-directed molecule optimization with 23 oracle functions.", "size": { "oracles": 23, "budget": 10000 }, "primary_paper": { "title": "Sample Efficiency Matters: A Benchmark for Practical Molecular Optimization", "authors": [ "Gao W", "Fu T", "Sun J", "Coley CW" ], "year": 2022, "doi": "10.48550/arXiv.2206.12411", "citations": 260 }, "official_url": "https://github.com/wenhao-gao/mol_opt", "github_url": "https://github.com/wenhao-gao/mol_opt", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2022-06", "last_updated": "2024-10", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 4, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Sample-efficiency focus exposed shortcomings of reward-maxing methods.", "related_benchmarks": [ "guacamol", "moses", "tdc-admet" ], "expert_ids": [ "wenhao-gao", "tianfan-fu", "connor-coley" ], "group_ids": [ "mit-csail", "coley-lab" ], "hosted_by": [ "tdc" ], "composite_score": 88.9 }, { "id": "cov-abdab", "name": "CoV-AbDab", "stages": [ "hit-id" ], "modalities": [ "biologic-mab" ], "task_types": [ "antibody-antigen-binding" ], "description": "Coronavirus antibody database \u2014 binding & neutralization annotations for SARS-CoV-2/MERS/etc.", "size": { "antibodies": 12000 }, "primary_paper": { "title": "CoV-AbDab: the coronavirus antibody database", "authors": [ "Raybould MIJ", "Kovaltsuk A", "Marks C", "Deane CM" ], "year": 2021, "doi": "10.1093/bioinformatics/btaa739", "citations": 220 }, "official_url": "https://opig.stats.ox.ac.uk/webapps/covabdab/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "CC-BY 4.0", "first_release": "2020", "last_updated": "2024-12", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 4, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Narrow modality but critical for pandemic-preparedness ML.", "related_benchmarks": [ "sabdab", "oas" ], "expert_ids": [ "matthew-raybould", "charlotte-deane" ], "group_ids": [ "oxford-opig" ], "hosted_by": [], "composite_score": 88.9 }, { "id": "pubchem-bioassay", "name": "PubChem BioAssay", "stages": [ "hit-id" ], "modalities": [ "small-molecule" ], "task_types": [ "bioactivity" ], "description": "NIH PubChem's screening assay repository.", "size": { "assays": 1700000, "compounds": 114000000 }, "primary_paper": { "title": "PubChem 2023 update", "authors": [ "Kim S", "Chen J", "Cheng T", "et al." ], "year": 2023, "doi": "10.1093/nar/gkac956", "citations": 1200 }, "official_url": "https://pubchem.ncbi.nlm.nih.gov/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "Public domain", "first_release": "2004", "last_updated": "2025-05", "rubric": { "rigor": 4, "coverage": 5, "maintenance": 5, "adoption": 5, "quality": 3, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Broadest HTS repository; quality heterogeneous.", "related_benchmarks": [ "chembl", "moleculenet" ], "expert_ids": [ "sunghwan-kim" ], "group_ids": [ "nih-ncbi" ], "hosted_by": [], "composite_score": 88.6 }, { "id": "polaris-admet", "name": "Polaris ADMET", "stages": [ "lead-id-admet" ], "modalities": [ "small-molecule" ], "task_types": [ "regression", "classification" ], "description": "Industry-contributed ADMET benchmarks on Polaris Hub (Novartis/AstraZeneca/Polaris SC curated endpoints).", "size": { "endpoints": 12, "molecules": 50000 }, "primary_paper": { "title": "Polaris: method comparison in drug discovery", "authors": [ "Wognum C", "Noutahi E", "Hsu J", "et al." ], "year": 2024, "doi": "N/A \u2014 working paper", "citations": 30 }, "official_url": "https://polarishub.io/benchmarks", "github_url": "https://github.com/polaris-hub/polaris", "leaderboard_url": "https://polarishub.io/benchmarks", "license": "Polaris Community", "first_release": "2023-10", "last_updated": "2025-04", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 5, "adoption": 3, "quality": 5, "accessibility": 4, "industry_relevance": 5 }, "flags": [], "notes": "Industry splits enforce blinded eval; highest industry relevance among ADMET benchmarks.", "related_benchmarks": [ "tdc-admet", "moleculenet" ], "expert_ids": [ "cas-wognum", "emmanuel-noutahi", "jonathan-hsu" ], "group_ids": [ "valence-labs", "recursion", "novartis-nibr", "astrazeneca" ], "hosted_by": [ "polaris" ], "composite_score": 88.4 }, { "id": "cz-virtual-cell-challenge", "name": "CZ Virtual Cell Challenge", "stages": [ "virtual-cell", "target-id" ], "modalities": [ "cross-modality" ], "task_types": [ "perturbation-prediction" ], "description": "Open challenge to predict transcriptomic responses to genetic & chemical perturbations across cell types, hosted by CZ Biohub.", "size": { "cells": 2000000, "perturbations": 300 }, "primary_paper": { "title": "CZ Virtual Cell Challenge (launch)", "authors": [ "CZI Science Team" ], "year": 2024, "doi": "N/A \u2014 consortium launch", "citations": 40 }, "official_url": "https://virtualcellchallenge.org/", "github_url": "https://github.com/czbiohub-sf", "leaderboard_url": "https://virtualcellchallenge.org/leaderboard", "license": "CC-BY", "first_release": "2024-11", "last_updated": "2025-09", "rubric": { "rigor": 4, "coverage": 5, "maintenance": 5, "adoption": 4, "quality": 5, "accessibility": 4, "industry_relevance": 4 }, "flags": [], "notes": "Gold standard-in-the-making for foundation-model era perturbation prediction. Hidden test \u2192 strong against leakage.", "related_benchmarks": [ "scperturb", "openproblems-perturbation", "perturbbench" ], "expert_ids": [ "stephen-quake", "ambrose-carr" ], "group_ids": [ "cz-biohub", "czi-science" ], "hosted_by": [ "czi-virtual-cell" ], "composite_score": 88.1 }, { "id": "pli-gpcr-suite", "name": "ISM Benchmarks: GPCRs (Insilico)", "stages": [ "hit-id", "lead-id-admet" ], "modalities": [ "small-molecule", "protein-general" ], "task_types": [ "binding-affinity" ], "description": "87-benchmark GPCR affinity suite on ScienceAIBench / InsilicoBench / DDB \u2014 kinome-scale coverage of class A/B/C GPCRs.", "size": { "benchmarks": 87, "gpcr_targets": 87 }, "primary_paper": { "title": "GPCR affinity benchmark methodology", "authors": "Insilico AI Team", "year": 2025, "doi": "N/A \u2014 portal", "citations": 8 }, "official_url": "https://scienceaibench.insilico.com/", "github_url": "N/A", "leaderboard_url": "https://scienceaibench.insilico.com/", "license": "CC-BY", "first_release": "2025-02", "last_updated": "2026-04", "rubric": { "rigor": 4, "coverage": 5, "maintenance": 5, "adoption": 3, "quality": 4, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Largest open GPCR affinity benchmark. Leaderboards test external frontier LLMs \u2014 not self-referential.", "related_benchmarks": [ "pdbbind", "chembl" ], "expert_ids": [ "alex-zhavoronkov", "alex-aliper", "alex-zhebrak" ], "group_ids": [ "insilico-medicine" ], "hosted_by": [ "insilico-scienceaibench", "insilico-insilicobench", "insilico-ddb" ], "composite_score": 87.6 }, { "id": "capri-benchmark", "name": "CAPRI Rounds", "stages": [ "hit-id" ], "modalities": [ "protein-general" ], "task_types": [ "protein-protein-docking" ], "description": "56 rounds of blind protein-protein (and peptide, ligand) complex prediction.", "size": { "rounds": 56, "targets": 300 }, "primary_paper": { "title": "Modeling protein-protein and protein-peptide complexes: CAPRI 6th edition", "authors": [ "Lensink MF", "Nadzirin N", "Velankar S", "Wodak SJ" ], "year": 2020, "doi": "10.1002/prot.25870", "citations": 200 }, "official_url": "https://www.ebi.ac.uk/pdbe/complex-pred/capri/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "Public", "first_release": "2001", "last_updated": "2024-11", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 4, "adoption": 4, "quality": 5, "accessibility": 4, "industry_relevance": 4 }, "flags": [], "notes": "Oldest PPI prediction benchmark.", "related_benchmarks": [ "pinder", "casp16" ], "expert_ids": [ "marc-lensink", "shoshana-wodak" ], "group_ids": [ "ebi", "ibm-brussels" ], "hosted_by": [ "capri" ], "composite_score": 86.3 }, { "id": "toxcast", "name": "ToxCast", "stages": [ "lead-id-admet", "ind-enabling" ], "modalities": [ "small-molecule" ], "task_types": [ "toxicity" ], "description": "EPA's in vitro toxicity screening dataset \u2014 ~700 assays \u00d7 ~9k chemicals.", "size": { "assays": 700, "compounds": 9000 }, "primary_paper": { "title": "The ToxCast Program for Prioritizing Toxicity Testing of Environmental Chemicals", "authors": [ "Dix DJ", "Houck KA", "Martin MT", "et al." ], "year": 2007, "doi": "10.1093/toxsci/kfm297", "citations": 1900 }, "official_url": "https://www.epa.gov/comptox-tools/exploring-toxcast-data", "github_url": "N/A", "leaderboard_url": "N/A", "license": "Public", "first_release": "2007", "last_updated": "2024-06", "rubric": { "rigor": 4, "coverage": 5, "maintenance": 4, "adoption": 4, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Regulatory-grade broad tox dataset.", "related_benchmarks": [ "tox21" ], "expert_ids": [ "richard-judson" ], "group_ids": [ "epa-ccte" ], "hosted_by": [ "tdc" ], "composite_score": 85.6 }, { "id": "targetbench-insilico", "name": "TargetBench (Insilico)", "stages": [ "target-id", "disease-modeling" ], "modalities": [ "cross-modality" ], "task_types": [ "target-prioritization" ], "description": "10 disease-area target identification benchmarks on ScienceAIBench/DDB (cancer, cardiovascular, endocrine/metabolic, fibrotic, inflammation/immunology, neuro, etc.).", "size": { "benchmarks": 10, "disease_areas": 10 }, "primary_paper": { "title": "TargetBench methodology", "authors": "Insilico AI Team", "year": 2025, "doi": "N/A \u2014 portal", "citations": 5 }, "official_url": "https://scienceaibench.insilico.com/", "github_url": "N/A", "leaderboard_url": "https://scienceaibench.insilico.com/", "license": "CC-BY", "first_release": "2025-03", "last_updated": "2026-04", "rubric": { "rigor": 4, "coverage": 4, "maintenance": 5, "adoption": 3, "quality": 4, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Disease-organized target ID benchmark \u2014 unique axis. Frontier LLM leaderboard.", "related_benchmarks": [ "open-targets", "depmap" ], "expert_ids": [ "alex-zhavoronkov", "alex-aliper" ], "group_ids": [ "insilico-medicine" ], "hosted_by": [ "insilico-scienceaibench", "insilico-ddb" ], "composite_score": 84.6 }, { "id": "ism-admet", "name": "ISM Benchmarks: ADMET (Insilico)", "stages": [ "lead-id-admet", "ind-enabling" ], "modalities": [ "small-molecule" ], "task_types": [ "admet-regression", "admet-classification" ], "description": "28-endpoint ADMET benchmark suite on ScienceAIBench/InsilicoBench/DDB.", "size": { "benchmarks": 28 }, "primary_paper": { "title": "ISM ADMET benchmark methodology", "authors": "Insilico AI Team", "year": 2025, "doi": "N/A \u2014 portal", "citations": 7 }, "official_url": "https://scienceaibench.insilico.com/", "github_url": "N/A", "leaderboard_url": "https://scienceaibench.insilico.com/", "license": "CC-BY", "first_release": "2025-02", "last_updated": "2026-04", "rubric": { "rigor": 4, "coverage": 4, "maintenance": 5, "adoption": 3, "quality": 4, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Broader endpoint coverage than TDC ADMET. Side-by-side with TDC mirror on DDB.", "related_benchmarks": [ "tdc-admet", "admet-ai" ], "expert_ids": [ "alex-zhavoronkov", "alex-aliper" ], "group_ids": [ "insilico-medicine" ], "hosted_by": [ "insilico-scienceaibench", "insilico-insilicobench", "insilico-ddb" ], "composite_score": 84.6 }, { "id": "cafa5", "name": "CAFA5", "stages": [ "target-id" ], "modalities": [ "protein-general" ], "task_types": [ "function-annotation" ], "description": "CAFA 5th edition \u2014 blind GO annotation eval (142k targets; Kaggle 2023, 1625 teams).", "size": { "targets": 142000, "go_terms": 43000 }, "primary_paper": { "title": "The CAFA challenge reports improved protein function prediction and new functional annotations for hundreds of genes through experimental screens", "authors": [ "Zhou N", "Jiang Y", "Bergquist TR", "et al." ], "year": 2019, "doi": "10.1186/s13059-019-1835-8", "citations": 380 }, "official_url": "https://biofunctionprediction.org/cafa/", "github_url": "N/A", "leaderboard_url": "https://biofunctionprediction.org/cafa/results", "license": "Public", "first_release": "2022", "last_updated": "2023-12", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 3, "adoption": 5, "quality": 4, "accessibility": 5, "industry_relevance": 3 }, "flags": [], "notes": "CAFA5 broke attendance records.", "related_benchmarks": [ "proteingym" ], "expert_ids": [ "predrag-radivojac", "iddo-friedberg" ], "group_ids": [ "cafa-consortium" ], "hosted_by": [ "cafa" ], "composite_score": 84.3 }, { "id": "molecule-ace", "name": "MoleculeACE", "stages": [ "lead-id-admet" ], "modalities": [ "small-molecule" ], "task_types": [ "regression", "activity-cliff" ], "description": "Benchmark testing model robustness on activity cliffs across 30 ChEMBL targets.", "size": { "targets": 30, "molecules": 48000 }, "primary_paper": { "title": "Exposing the Limitations of Molecular Machine Learning with Activity Cliffs", "authors": [ "van Tilborg D", "Alenicheva A", "Grisoni F" ], "year": 2022, "doi": "10.1021/acs.jcim.2c01073", "citations": 180 }, "official_url": "https://github.com/molML/MoleculeACE", "github_url": "https://github.com/molML/MoleculeACE", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2022-11", "last_updated": "2024-05", "rubric": { "rigor": 5, "coverage": 3, "maintenance": 3, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Critical stress-test for generalization; exposed GNN weaknesses.", "related_benchmarks": [ "tdc-admet", "moleculenet" ], "expert_ids": [ "francesca-grisoni", "derek-van-tilborg" ], "group_ids": [ "tue-eindhoven" ], "hosted_by": [], "composite_score": 83.3 }, { "id": "matbench", "name": "MatBench", "stages": [ "developmental-candidate" ], "modalities": [ "cross-modality" ], "task_types": [ "materials-property" ], "description": "Materials property prediction benchmark suite (used for some chemistry-adjacent ML).", "size": { "tasks": 13, "samples": 132000 }, "primary_paper": { "title": "Benchmarking materials property prediction methods: the Matbench test set and Automatminer reference algorithm", "authors": [ "Dunn A", "Wang Q", "Ganose A", "Dopp D", "Jain A" ], "year": 2020, "doi": "10.1038/s41524-020-00406-3", "citations": 380 }, "official_url": "https://matbench.materialsproject.org/", "github_url": "https://github.com/materialsproject/matbench", "leaderboard_url": "https://matbench.materialsproject.org/Leaderboards%20Per-Task/", "license": "MIT", "first_release": "2020", "last_updated": "2025-02", "rubric": { "rigor": 5, "coverage": 3, "maintenance": 4, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 3 }, "flags": [], "notes": "Materials-science benchmark; relevant for formulation / co-crystal work.", "related_benchmarks": [], "expert_ids": [ "anubhav-jain", "alex-dunn" ], "group_ids": [ "materials-project", "lbl" ], "hosted_by": [ "insilico-scienceaibench", "insilico-ddb" ], "composite_score": 83.3 }, { "id": "offsides-twosides", "name": "OffSides / TWOSIDES", "stages": [ "post-market-rwe" ], "modalities": [ "small-molecule" ], "task_types": [ "adverse-event", "ddi" ], "description": "Statistically corrected single- and pair-wise drug adverse events derived from FAERS.", "size": { "offsides_signals": 438801, "twosides_combos": 870000 }, "primary_paper": { "title": "Data-driven prediction of drug effects and interactions", "authors": [ "Tatonetti NP", "Ye PP", "Daneshjou R", "Altman RB" ], "year": 2012, "doi": "10.1126/scitranslmed.3003377", "citations": 1100 }, "official_url": "http://tatonettilab.org/offsides/", "github_url": "https://github.com/tatonetti-lab/nsides-release", "leaderboard_url": "N/A", "license": "CC-BY 4.0", "first_release": "2012", "last_updated": "2023-09", "rubric": { "rigor": 4, "coverage": 4, "maintenance": 3, "adoption": 5, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Key benchmark for DDI + adverse event ML.", "related_benchmarks": [ "sider", "faers-bench" ], "expert_ids": [ "nick-tatonetti", "russ-altman" ], "group_ids": [ "tatonetti-lab" ], "hosted_by": [ "faers", "tdc" ], "composite_score": 83.0 }, { "id": "clinbench-quarterly", "name": "ClinBench Quarterly (Insilico)", "stages": [ "phase-2", "phase-3", "clinical-development" ], "modalities": [ "cross-modality" ], "task_types": [ "trial-outcome-prediction" ], "description": "Quarterly-refreshed clinical-trial outcome benchmark on ScienceAIBench / InsilicoBench / DDB (25 tasks).", "size": { "tasks": 25, "refresh_cadence_months": 3 }, "primary_paper": { "title": "ClinBench methodology note", "authors": "Insilico AI Team", "year": 2025, "doi": "N/A \u2014 portal", "citations": 10 }, "official_url": "https://scienceaibench.insilico.com/", "github_url": "N/A", "leaderboard_url": "https://scienceaibench.insilico.com/", "license": "CC-BY", "first_release": "2025-01", "last_updated": "2026-04", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 5, "adoption": 3, "quality": 4, "accessibility": 5, "industry_relevance": 5 }, "flags": [], "notes": "Benchmark refresh cadence beats all academic trial outcome benchmarks. Leaderboards test frontier LLMs against quarterly-updated splits.", "related_benchmarks": [ "hint-trialbench", "top-benchmark" ], "expert_ids": [ "alex-zhavoronkov", "alex-aliper" ], "group_ids": [ "insilico-medicine" ], "hosted_by": [ "insilico-scienceaibench", "insilico-insilicobench", "insilico-ddb" ], "composite_score": 81.5 }, { "id": "dockstring", "name": "DOCKSTRING", "stages": [ "hit-id" ], "modalities": [ "small-molecule" ], "task_types": [ "docking" ], "description": "260k ligands docked to 58 targets with AutoDock Vina; reproducible docking benchmark.", "size": { "ligands": 260155, "targets": 58 }, "primary_paper": { "title": "DOCKSTRING: easy molecular docking yields better benchmarks for ligand design", "authors": [ "Garc\u00eda-Orteg\u00f3n M", "Simm GNC", "Tripp AJ", "et al." ], "year": 2022, "doi": "10.1021/acs.jcim.1c01334", "citations": 240 }, "official_url": "https://dockstring.github.io/", "github_url": "https://github.com/dockstring/dockstring", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2022-02", "last_updated": "2024-07", "rubric": { "rigor": 5, "coverage": 4, "maintenance": 3, "adoption": 4, "quality": 4, "accessibility": 5, "industry_relevance": 3 }, "flags": [], "notes": "Vina scores are a proxy; not a replacement for wet assays.", "related_benchmarks": [ "lit-pcba", "pdbbind", "casf-2016" ], "expert_ids": [ "jose-miguel-hernandez-lobato" ], "group_ids": [ "cambridge-ml" ], "hosted_by": [], "composite_score": 81.3 }, { "id": "disgenet", "name": "DisGeNET", "stages": [ "disease-modeling", "target-id" ], "modalities": [ "cross-modality" ], "task_types": [ "gene-disease-association" ], "description": "Large collection of gene-disease associations (text-mining + curation + repositories).", "size": { "associations": 1134942, "genes": 21671, "diseases": 30170 }, "primary_paper": { "title": "The DisGeNET knowledge platform for disease genomics: 2019 update", "authors": [ "Pi\u00f1ero J", "Ram\u00edrez-Anguita JM", "Sa\u00fcch-Pitarch J", "et al." ], "year": 2020, "doi": "10.1093/nar/gkz1021", "citations": 1400 }, "official_url": "https://www.disgenet.com/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "CC-BY-NC / commercial tier", "first_release": "2010", "last_updated": "2024-11", "rubric": { "rigor": 4, "coverage": 5, "maintenance": 4, "adoption": 5, "quality": 3, "accessibility": 3, "industry_relevance": 4 }, "flags": [ "license-gated-commercial" ], "notes": "Commercial license required for industry. Text-mining noise limits quality.", "related_benchmarks": [ "open-targets", "primekg" ], "expert_ids": [ "laura-furlong" ], "group_ids": [ "medbioinformatics" ], "hosted_by": [ "tdc" ], "composite_score": 81.0 }, { "id": "lit-pcba", "name": "LIT-PCBA", "stages": [ "hit-id" ], "modalities": [ "small-molecule" ], "task_types": [ "virtual-screening" ], "description": "Literature-curated dose-response PubChem BioAssay set; 15 targets with unbiased actives/inactives.", "size": { "targets": 15, "actives": 7844, "inactives": 2533936 }, "primary_paper": { "title": "LIT-PCBA: An Unbiased Data Set for Machine Learning and Virtual Screening", "authors": [ "Tran-Nguyen VK", "Jacquemard C", "Rognan D" ], "year": 2020, "doi": "10.1021/acs.jcim.0c00155", "citations": 310 }, "official_url": "https://drugdesign.unistra.fr/LIT-PCBA/", "github_url": "https://github.com/ViktorTran-Nguyen/LIT-PCBA", "leaderboard_url": "N/A", "license": "CC-BY", "first_release": "2020-03", "last_updated": "2023-06", "rubric": { "rigor": 5, "coverage": 3, "maintenance": 2, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Much fairer than DUD-E; small target count limits coverage.", "related_benchmarks": [ "dude", "dekois", "dockstring" ], "expert_ids": [ "didier-rognan" ], "group_ids": [ "strasbourg-chemo" ], "hosted_by": [], "composite_score": 80.8 }, { "id": "flip", "name": "FLIP", "stages": [ "target-id", "developmental-candidate" ], "modalities": [ "protein-general" ], "task_types": [ "fitness-prediction" ], "description": "Fitness landscape inference benchmarks with realistic train/test splits (AAV, GB1, Meltome, SCL, Bind).", "size": { "landscapes": 5, "splits": 15 }, "primary_paper": { "title": "FLIP: Benchmark tasks in fitness landscape inference for proteins", "authors": [ "Dallago C", "Mou J", "Johnston KE", "et al." ], "year": 2021, "doi": "10.48550/arXiv.2112.06661", "citations": 120 }, "official_url": "https://benchmark.protein.properties/", "github_url": "https://github.com/J-SNACKKB/FLIP", "leaderboard_url": "N/A", "license": "CC-BY 4.0", "first_release": "2021-12", "last_updated": "2024-05", "rubric": { "rigor": 5, "coverage": 3, "maintenance": 3, "adoption": 4, "quality": 5, "accessibility": 5, "industry_relevance": 3 }, "flags": [], "notes": "Complements ProteinGym (smaller but carefully designed splits).", "related_benchmarks": [ "proteingym" ], "expert_ids": [ "burkhard-rost", "mohammed-alquraishi", "christian-dallago" ], "group_ids": [ "rostlab-tum", "alquraishi-lab" ], "hosted_by": [ "flip" ], "composite_score": 80.8 }, { "id": "cptac-proteogenomic", "name": "CPTAC Proteogenomic Benchmarks", "stages": [ "disease-modeling", "target-id", "phase-2" ], "modalities": [ "cross-modality" ], "task_types": [ "biomarker-discovery" ], "description": "Proteogenomic benchmarks across 10 tumor types \u2014 drives DREAM proteogenomic challenges.", "size": { "samples": 1600, "tumor_types": 10, "omics_layers": 6 }, "primary_paper": { "title": "Proteogenomic Characterization of Cancer Types (CPTAC overview)", "authors": [ "Zhang B", "Wang J", "Wang X", "et al." ], "year": 2014, "doi": "10.1038/nature13438", "citations": 1200 }, "official_url": "https://proteomics.cancer.gov/programs/cptac", "github_url": "https://github.com/PayneLab/cptac", "leaderboard_url": "N/A", "license": "dbGaP / public", "first_release": "2011", "last_updated": "2025-03", "rubric": { "rigor": 5, "coverage": 3, "maintenance": 4, "adoption": 4, "quality": 5, "accessibility": 3, "industry_relevance": 4 }, "flags": [], "notes": "Deep integrative oncology data.", "related_benchmarks": [ "depmap" ], "expert_ids": [ "henry-rodriguez", "amanda-paulovich", "bing-zhang" ], "group_ids": [ "nci-cptac" ], "hosted_by": [ "cptac" ], "composite_score": 80.8 }, { "id": "guacamol", "name": "GuacaMol", "stages": [ "lead-id-admet", "developmental-candidate" ], "modalities": [ "small-molecule" ], "task_types": [ "molecule-generation" ], "description": "Goal-directed + distribution-learning benchmarks for molecular generative models.", "size": { "tasks": 20, "train_set": 1600000 }, "primary_paper": { "title": "GuacaMol: Benchmarking Models for de Novo Molecular Design", "authors": [ "Brown N", "Fiscato M", "Segler MHS", "Vaucher AC" ], "year": 2019, "doi": "10.1021/acs.jcim.8b00839", "citations": 820 }, "official_url": "https://www.benevolent.com/guacamol", "github_url": "https://github.com/BenevolentAI/guacamol", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2019-03", "last_updated": "2022-07", "rubric": { "rigor": 4, "coverage": 4, "maintenance": 2, "adoption": 5, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "First-generation generative benchmark; largely superseded by PMO for goal-directed.", "related_benchmarks": [ "moses", "pmo" ], "expert_ids": [ "marwin-segler", "nathan-brown" ], "group_ids": [ "benevolent-ai" ], "hosted_by": [ "moleculenet", "tdc" ], "composite_score": 80.5 }, { "id": "pksim", "name": "Open Systems Pharmacology / PK-Sim", "stages": [ "phase-1", "ind-enabling" ], "modalities": [ "small-molecule", "biologic-mab" ], "task_types": [ "pbpk-validation" ], "description": "OSP Suite \u2014 open PBPK/QSP models and validation sets.", "size": { "models": 100, "validated_compounds": 50 }, "primary_paper": { "title": "The Open Systems Pharmacology Suite: a new era in PBPK modeling", "authors": [ "Lippert J", "Burghaus R", "Edginton A", "et al." ], "year": 2019, "doi": "10.1002/psp4.12386", "citations": 180 }, "official_url": "https://www.open-systems-pharmacology.org/", "github_url": "https://github.com/Open-Systems-Pharmacology", "leaderboard_url": "N/A", "license": "GPL-2.0", "first_release": "2017", "last_updated": "2025-01", "rubric": { "rigor": 5, "coverage": 3, "maintenance": 4, "adoption": 3, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Open alternative to Simcyp.", "related_benchmarks": [ "simcyp-validation" ], "expert_ids": [ "andrea-edginton", "joerg-lippert" ], "group_ids": [ "osp-consortium" ], "hosted_by": [], "composite_score": 80.3 }, { "id": "admet-ai", "name": "ADMET-AI", "stages": [ "lead-id-admet" ], "modalities": [ "small-molecule" ], "task_types": [ "regression", "classification" ], "description": "Graph-based ADMET prediction + benchmark on 41 endpoints leveraging ChEMBL + TDC.", "size": { "endpoints": 41, "molecules": 90000 }, "primary_paper": { "title": "ADMET-AI: a machine learning ADMET platform for evaluation of large-scale chemical libraries", "authors": [ "Swanson K", "Walther P", "Leitz J", "et al." ], "year": 2023, "doi": "10.1093/bioinformatics/btae416", "citations": 85 }, "official_url": "https://admet.ai.greenstonebio.com/", "github_url": "https://github.com/swansonk14/admet_ai", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2023-11", "last_updated": "2024-12", "rubric": { "rigor": 4, "coverage": 4, "maintenance": 4, "adoption": 3, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Strong baselines + web tool; builds on TDC.", "related_benchmarks": [ "tdc-admet", "moleculenet" ], "expert_ids": [ "kyle-swanson", "regina-barzilay" ], "group_ids": [ "mit-csail", "barzilay-lab" ], "hosted_by": [], "composite_score": 79.5 }, { "id": "ames", "name": "AMES (mutagenicity)", "stages": [ "ind-enabling", "lead-id-admet" ], "modalities": [ "small-molecule" ], "task_types": [ "toxicity-classification" ], "description": "AMES bacterial mutagenicity benchmark \u2014 standard gentox endpoint.", "size": { "molecules": 7278 }, "primary_paper": { "title": "Therapeutics Data Commons", "authors": [ "Huang K", "Fu T", "et al." ], "year": 2021, "doi": "10.48550/arXiv.2102.09548", "citations": 620 }, "official_url": "https://tdcommons.ai/single_pred_tasks/tox/#ames", "github_url": "https://github.com/mims-harvard/TDC", "leaderboard_url": "https://tdcommons.ai/benchmark/admet_group/22ames/", "license": "MIT", "first_release": "2021-02", "last_updated": "2025-01", "rubric": { "rigor": 4, "coverage": 2, "maintenance": 4, "adoption": 5, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Core gentox endpoint.", "related_benchmarks": [ "tdc-admet", "tox21" ], "expert_ids": [ "kexin-huang" ], "group_ids": [ "tdc" ], "hosted_by": [ "tdc", "insilico-ddb" ], "composite_score": 79.5 }, { "id": "polaris-biologics", "name": "Polaris Biologics (Polyreactivity / SEC / Tm)", "stages": [ "developmental-candidate" ], "modalities": [ "biologic-mab" ], "task_types": [ "developability" ], "description": "Polaris Hub biologics benchmarks: polyreactivity, SEC-SMAC, Tm + titer.", "size": { "tasks": 6, "antibodies": 2000 }, "primary_paper": { "title": "Polaris biologics method comparison", "authors": "Wognum C et al.", "year": 2024, "doi": "N/A", "citations": 15 }, "official_url": "https://polarishub.io/benchmarks", "github_url": "https://github.com/polaris-hub/polaris", "leaderboard_url": "https://polarishub.io/benchmarks", "license": "Polaris Community", "first_release": "2024-04", "last_updated": "2025-03", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 5, "adoption": 3, "quality": 4, "accessibility": 4, "industry_relevance": 5 }, "flags": [], "notes": "Industry-donated; growing.", "related_benchmarks": [ "polaris-admet" ], "expert_ids": [ "cas-wognum" ], "group_ids": [ "valence-labs", "recursion" ], "hosted_by": [ "polaris", "insilico-scienceaibench", "insilico-ddb" ], "composite_score": 79.0 }, { "id": "moleculenet", "name": "MoleculeNet", "stages": [ "lead-id-admet", "hit-id" ], "modalities": [ "small-molecule" ], "task_types": [ "regression", "classification", "quantum" ], "description": "Multi-task molecular ML benchmark covering quantum, physical, biophysical, physiological properties.", "size": { "tasks": 17, "molecules": 700000 }, "primary_paper": { "title": "MoleculeNet: A Benchmark for Molecular Machine Learning", "authors": [ "Wu Z", "Ramsundar B", "Feinberg EN", "et al." ], "year": 2018, "doi": "10.1039/C7SC02664A", "citations": 3600 }, "official_url": "https://moleculenet.org/", "github_url": "https://github.com/deepchem/deepchem", "leaderboard_url": "https://moleculenet.org/full-results", "license": "MIT", "first_release": "2018-03", "last_updated": "2023-11", "rubric": { "rigor": 4, "coverage": 4, "maintenance": 2, "adoption": 5, "quality": 3, "accessibility": 5, "industry_relevance": 4 }, "flags": [ "data-leakage-known" ], "notes": "Widely cited (3600+); aging splits with known scaffold leakage.", "related_benchmarks": [ "tdc-admet", "polaris-admet" ], "expert_ids": [ "bharath-ramsundar", "vijay-pande" ], "group_ids": [ "pande-lab", "deepchem" ], "hosted_by": [ "moleculenet", "deepchem", "papers-with-code-drug" ], "composite_score": 78.0 }, { "id": "uspto-retrosyn", "name": "USPTO-50K / USPTO-MIT (Retrosynthesis)", "stages": [ "lead-id-admet", "developmental-candidate" ], "modalities": [ "small-molecule" ], "task_types": [ "retrosynthesis", "reaction-prediction" ], "description": "Reactions extracted from USPTO patents; standard retrosynthesis/forward-reaction benchmark.", "size": { "reactions": 1800000, "canonical_50k": 50037 }, "primary_paper": { "title": "Neural Sequence-to-Sequence Models for Retrosynthesis Prediction", "authors": [ "Liu B", "Ramsundar B", "Kawthekar P", "et al." ], "year": 2017, "doi": "10.1021/acscentsci.7b00303", "citations": 520 }, "official_url": "https://github.com/Hanjun-Dai/GLN", "github_url": "https://github.com/Hanjun-Dai/GLN", "leaderboard_url": "N/A", "license": "Public", "first_release": "2017", "last_updated": "2023", "rubric": { "rigor": 4, "coverage": 4, "maintenance": 2, "adoption": 5, "quality": 3, "accessibility": 5, "industry_relevance": 4 }, "flags": [ "data-leakage-known" ], "notes": "Known leakage across canonical splits; use time-split or ORD for fairer eval.", "related_benchmarks": [ "open-reaction-database", "chemrxiv-reactions" ], "expert_ids": [ "bharath-ramsundar", "connor-coley" ], "group_ids": [ "mit-csail", "coley-lab" ], "hosted_by": [ "tdc", "papers-with-code-drug" ], "composite_score": 78.0 }, { "id": "tox21", "name": "Tox21", "stages": [ "lead-id-admet", "ind-enabling" ], "modalities": [ "small-molecule" ], "task_types": [ "toxicity-classification" ], "description": "US Tox21 program HTS data on 12 nuclear receptor / stress response assays.", "size": { "compounds": 10000, "assays": 12 }, "primary_paper": { "title": "Tox21 Challenge to Build Predictive Models of Nuclear Receptor and Stress Response Pathways", "authors": [ "Huang R", "Xia M", "et al." ], "year": 2016, "doi": "10.3389/fenvs.2015.00085", "citations": 1100 }, "official_url": "https://tripod.nih.gov/tox21/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "Public", "first_release": "2014", "last_updated": "2017", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 2, "adoption": 5, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Field-standard tox benchmark; endpoint count small vs modern suites.", "related_benchmarks": [ "toxcast", "moleculenet" ], "expert_ids": [ "ruili-huang" ], "group_ids": [ "ncats" ], "hosted_by": [ "moleculenet", "tdc" ], "composite_score": 77.5 }, { "id": "iglm-bench", "name": "IgLM / AntiBERTa benchmarks", "stages": [ "hit-id", "developmental-candidate" ], "modalities": [ "biologic-mab" ], "task_types": [ "antibody-generation", "liability-prediction" ], "description": "Antibody LM eval \u2014 paratope prediction, CDR generation, developability.", "size": { "sequences": 600000000, "tasks": 6 }, "primary_paper": { "title": "Generative language models for antibody design", "authors": [ "Shuai RW", "Ruffolo JA", "Gray JJ" ], "year": 2023, "doi": "10.1016/j.cels.2023.07.001", "citations": 140 }, "official_url": "https://github.com/Graylab/IgLM", "github_url": "https://github.com/Graylab/IgLM", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2022", "last_updated": "2024-08", "rubric": { "rigor": 4, "coverage": 4, "maintenance": 3, "adoption": 4, "quality": 4, "accessibility": 4, "industry_relevance": 4 }, "flags": [], "notes": "Moves toward true developability benchmarks.", "related_benchmarks": [ "oas", "sabdab" ], "expert_ids": [ "jeffrey-gray", "richard-shuai" ], "group_ids": [ "jhu-gray-lab" ], "hosted_by": [], "composite_score": 77.5 }, { "id": "geneformer-bench", "name": "Geneformer Eval", "stages": [ "virtual-cell" ], "modalities": [ "cross-modality" ], "task_types": [ "cell-type-annotation", "gene-dosage" ], "description": "Evaluation of Geneformer (30M transformer pretrained on single-cell corpus) on zero-shot & few-shot downstream tasks.", "size": { "cells": 30000000, "tasks": 7 }, "primary_paper": { "title": "Transfer learning enables predictions in network biology", "authors": [ "Theodoris CV", "Xiao L", "Chopra A", "et al." ], "year": 2023, "doi": "10.1038/s41586-023-06139-9", "citations": 550 }, "official_url": "https://huggingface.co/ctheodoris/Geneformer", "github_url": "https://huggingface.co/ctheodoris/Geneformer", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2023-06", "last_updated": "2024-11", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 4, "adoption": 4, "quality": 4, "accessibility": 4, "industry_relevance": 4 }, "flags": [ "self_referential" ], "notes": "Author-led eval; still widely re-run on OpenProblems tasks.", "related_benchmarks": [ "openproblems-perturbation", "scgpt-bench" ], "expert_ids": [ "christina-theodoris" ], "group_ids": [ "broad-institute" ], "hosted_by": [], "composite_score": 77.0 }, { "id": "tdc-drug-syn", "name": "TDC DrugSyn (OncoPolyPharm + DrugComb_NCI60)", "stages": [ "developmental-candidate", "lead-id-admet" ], "modalities": [ "small-molecule" ], "task_types": [ "drug-synergy" ], "description": "Drug combination synergy benchmark (NCI ALMANAC + OncoPolyPharmacology).", "size": { "combos": 23000, "cell_lines": 60 }, "primary_paper": { "title": "Therapeutics Data Commons", "authors": [ "Huang K", "Fu T", "et al." ], "year": 2021, "doi": "10.48550/arXiv.2102.09548", "citations": 620 }, "official_url": "https://tdcommons.ai/multi_pred_tasks/drugsyn/", "github_url": "https://github.com/mims-harvard/TDC", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2021-02", "last_updated": "2024-12", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 4, "adoption": 4, "quality": 4, "accessibility": 5, "industry_relevance": 3 }, "flags": [], "notes": "Important for combination therapy design.", "related_benchmarks": [ "tdc-admet", "depmap" ], "expert_ids": [ "kexin-huang", "marinka-zitnik" ], "group_ids": [ "tdc", "zitnik-lab" ], "hosted_by": [ "tdc" ], "composite_score": 77.0 }, { "id": "pkpd-obach", "name": "Obach PK Dataset", "stages": [ "phase-1", "ind-enabling", "lead-id-admet" ], "modalities": [ "small-molecule" ], "task_types": [ "pk-regression" ], "description": "Obach human PK dataset (t1/2, VDss, CL) \u2014 standard human-PK ML benchmark.", "size": { "compounds": 1338, "endpoints": 3 }, "primary_paper": { "title": "Trend analysis of a database of intravenous pharmacokinetic parameters in humans for 670 drug compounds", "authors": [ "Obach RS", "Lombardo F", "Waters NJ" ], "year": 2008, "doi": "10.1124/dmd.108.020479", "citations": 820 }, "official_url": "https://tdcommons.ai/single_pred_tasks/adme/#half-life-obach-et-al", "github_url": "https://github.com/mims-harvard/TDC", "leaderboard_url": "https://tdcommons.ai/benchmark/admet_group/", "license": "MIT", "first_release": "2008", "last_updated": "2024-06", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 3, "adoption": 4, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Small but highest-quality human-PK dataset.", "related_benchmarks": [ "tdc-admet" ], "expert_ids": [ "scott-obach", "franco-lombardo" ], "group_ids": [ "pfizer" ], "hosted_by": [ "tdc", "insilico-ddb" ], "composite_score": 77.0 }, { "id": "hint-trialbench", "name": "HINT / TrialBench", "stages": [ "phase-2", "phase-3", "clinical-development" ], "modalities": [ "cross-modality" ], "task_types": [ "trial-outcome-prediction" ], "description": "Clinical trial outcome prediction benchmarks built on ClinicalTrials.gov (17-21k trials).", "size": { "trials": 21000, "drugs": 12000, "diseases": 5000 }, "primary_paper": { "title": "Hierarchical Interaction Network for Clinical Trial Outcome Prediction", "authors": [ "Fu T", "Huang K", "Xiao C", "Glass L", "Sun J" ], "year": 2022, "doi": "10.1016/j.patter.2022.100445", "citations": 200 }, "official_url": "https://github.com/futianfan/clinical-trial-outcome-prediction", "github_url": "https://github.com/futianfan/clinical-trial-outcome-prediction", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2022", "last_updated": "2024-07", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 4, "adoption": 3, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Limited by ClinicalTrials.gov quality.", "related_benchmarks": [ "top-benchmark", "ctod" ], "expert_ids": [ "tianfan-fu", "jimeng-sun", "marinka-zitnik" ], "group_ids": [ "fu-lab", "sun-lab-gatech", "zitnik-lab" ], "hosted_by": [ "trialbench", "tdc" ], "composite_score": 76.5 }, { "id": "top-benchmark", "name": "Trial Outcome Prediction (TOP)", "stages": [ "phase-3", "clinical-development" ], "modalities": [ "cross-modality" ], "task_types": [ "trial-outcome-prediction" ], "description": "Benchmarks for predicting Phase 1-3 trial outcomes from structured + text features.", "size": { "trials": 17000, "phases": 3 }, "primary_paper": { "title": "Artificial intelligence for clinical trial design", "authors": [ "Harrer S", "Shah P", "Antony B", "Hu J" ], "year": 2019, "doi": "10.1016/j.tips.2019.05.005", "citations": 520 }, "official_url": "https://github.com/futianfan/clinical-trial-outcome-prediction", "github_url": "https://github.com/futianfan/clinical-trial-outcome-prediction", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2022", "last_updated": "2024", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 4, "adoption": 3, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Often reported alongside HINT.", "related_benchmarks": [ "hint-trialbench" ], "expert_ids": [ "tianfan-fu", "jimeng-sun" ], "group_ids": [ "fu-lab", "sun-lab-gatech" ], "hosted_by": [ "trialbench" ], "composite_score": 76.5 }, { "id": "casf-2016", "name": "CASF-2016", "stages": [ "hit-id" ], "modalities": [ "small-molecule", "protein-general" ], "task_types": [ "scoring-function-eval" ], "description": "Comparative Assessment of Scoring Functions \u2014 scoring, ranking, docking, screening power tests.", "size": { "complexes": 285, "decoy_poses": 28500 }, "primary_paper": { "title": "Comparative Assessment of Scoring Functions: The CASF-2016 Update", "authors": [ "Su M", "Yang Q", "Du Y", "et al." ], "year": 2019, "doi": "10.1021/acs.jcim.8b00545", "citations": 430 }, "official_url": "http://www.pdbbind.org.cn/casf.php", "github_url": "N/A", "leaderboard_url": "N/A", "license": "Academic-only", "first_release": "2016", "last_updated": "2019", "rubric": { "rigor": 5, "coverage": 3, "maintenance": 2, "adoption": 5, "quality": 4, "accessibility": 3, "industry_relevance": 4 }, "flags": [], "notes": "Authoritative scoring-power eval; update cadence slow.", "related_benchmarks": [ "pdbbind", "posebusters", "plinder" ], "expert_ids": [ "renxiao-wang" ], "group_ids": [ "simm-shanghai" ], "hosted_by": [ "pdbbind-casf" ], "composite_score": 76.2 }, { "id": "pdbbind", "name": "PDBbind", "stages": [ "hit-id", "lead-id-admet" ], "modalities": [ "small-molecule", "protein-general" ], "task_types": [ "binding-affinity" ], "description": "Curated PDB complexes with experimental binding affinities \u2014 de facto standard for ML affinity prediction.", "size": { "complexes": 23500, "general_set": 19443, "refined_set": 5316 }, "primary_paper": { "title": "The PDBbind Database: Methodologies and Updates", "authors": [ "Wang R", "Fang X", "Lu Y", "Wang S" ], "year": 2005, "doi": "10.1021/jm048957q", "citations": 2100 }, "official_url": "http://www.pdbbind.org.cn/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "Academic-only", "first_release": "2004", "last_updated": "2022-01", "rubric": { "rigor": 4, "coverage": 5, "maintenance": 2, "adoption": 5, "quality": 3, "accessibility": 3, "industry_relevance": 4 }, "flags": [ "data-leakage-known" ], "notes": "Scaffold/temporal leakage well-documented. Pair with CASF + LeakyPDB.", "related_benchmarks": [ "casf-2016", "plinder", "posebusters" ], "expert_ids": [ "renxiao-wang" ], "group_ids": [ "simm-shanghai" ], "hosted_by": [ "pdbbind-casf" ], "composite_score": 75.9 }, { "id": "sider", "name": "SIDER", "stages": [ "post-market-rwe", "ind-enabling" ], "modalities": [ "small-molecule" ], "task_types": [ "adverse-event" ], "description": "Drug-side effect associations mined from FDA labels.", "size": { "drugs": 1430, "side_effect_pairs": 139000 }, "primary_paper": { "title": "The SIDER database of drugs and side effects", "authors": [ "Kuhn M", "Letunic I", "Jensen LJ", "Bork P" ], "year": 2016, "doi": "10.1093/nar/gkv1075", "citations": 1700 }, "official_url": "http://sideeffects.embl.de/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "CC-BY-NC-SA 4.0", "first_release": "2010", "last_updated": "2016", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 2, "adoption": 5, "quality": 3, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Aging but still widely used. TWOSIDES/OffSides offer newer signals.", "related_benchmarks": [ "offsides-twosides", "faers-bench" ], "expert_ids": [ "michael-kuhn", "peer-bork" ], "group_ids": [ "embl-bork" ], "hosted_by": [ "faers" ], "composite_score": 74.9 }, { "id": "tape", "name": "TAPE", "stages": [ "target-id", "developmental-candidate" ], "modalities": [ "protein-general" ], "task_types": [ "protein-ml" ], "description": "Tasks Assessing Protein Embeddings \u2014 5 tasks (secondary structure, contact, fluorescence, stability, homology).", "size": { "tasks": 5 }, "primary_paper": { "title": "Evaluating Protein Transfer Learning with TAPE", "authors": [ "Rao R", "Bhattacharya N", "Thomas N", "Dai Y", "Liu P", "Canny J", "Abbeel P", "Song YS" ], "year": 2019, "doi": "10.48550/arXiv.1906.08230", "citations": 950 }, "official_url": "https://github.com/songlab-cal/tape", "github_url": "https://github.com/songlab-cal/tape", "leaderboard_url": "N/A", "license": "BSD-3", "first_release": "2019", "last_updated": "2022", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 2, "adoption": 5, "quality": 4, "accessibility": 5, "industry_relevance": 3 }, "flags": [ "deprecated-recommend-replace" ], "notes": "Historically important; largely superseded by ProteinGym/FLIP for fitness and by PEER for broader tasks.", "related_benchmarks": [ "proteingym", "peer" ], "expert_ids": [ "roshan-rao", "pieter-abbeel", "yun-song" ], "group_ids": [ "uc-berkeley" ], "hosted_by": [ "papers-with-code-drug" ], "composite_score": 74.9 }, { "id": "simcyp-validation", "name": "Simcyp Validation Sets", "stages": [ "phase-1", "phase-2", "ind-enabling" ], "modalities": [ "small-molecule" ], "task_types": [ "pbpk-validation" ], "description": "PBPK validation datasets used by Simcyp/Certara community (DDI, pediatric, renal impairment).", "size": { "scenarios": 100 }, "primary_paper": { "title": "Physiologically based pharmacokinetic modeling: Methods and applications in pharmacotherapy", "authors": [ "Rostami-Hodjegan A", "Tucker GT" ], "year": 2007, "doi": "10.1038/nrd2173", "citations": 780 }, "official_url": "https://www.certara.com/software/simcyp/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "Proprietary", "first_release": "2001", "last_updated": "2024", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 4, "adoption": 4, "quality": 4, "accessibility": 2, "industry_relevance": 5 }, "flags": [ "license-gated-commercial" ], "notes": "Industry gold standard but proprietary. Open benchmarks exist via OSP Suite.", "related_benchmarks": [ "pksim" ], "expert_ids": [ "amin-rostami-hodjegan" ], "group_ids": [ "certara" ], "hosted_by": [], "composite_score": 74.4 }, { "id": "peer", "name": "PEER", "stages": [ "target-id", "developmental-candidate" ], "modalities": [ "protein-general" ], "task_types": [ "protein-ml" ], "description": "PEER \u2014 14 protein property prediction tasks across structure/function/interaction.", "size": { "tasks": 14 }, "primary_paper": { "title": "PEER: A Comprehensive and Multi-Task Benchmark for Protein Sequence Understanding", "authors": [ "Xu M", "Yuan X", "Miret S", "Tang J" ], "year": 2022, "doi": "10.48550/arXiv.2206.02096", "citations": 130 }, "official_url": "https://github.com/DeepGraphLearning/PEER_Benchmark", "github_url": "https://github.com/DeepGraphLearning/PEER_Benchmark", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2022", "last_updated": "2024", "rubric": { "rigor": 4, "coverage": 4, "maintenance": 3, "adoption": 3, "quality": 4, "accessibility": 5, "industry_relevance": 3 }, "flags": [], "notes": "Broader than TAPE, tighter than ProteinGym; good middle ground.", "related_benchmarks": [ "tape", "proteingym", "flip" ], "expert_ids": [ "jian-tang", "minghao-xu" ], "group_ids": [ "mila-quebec" ], "hosted_by": [ "papers-with-code-drug" ], "composite_score": 74.4 }, { "id": "clawbio-bench", "name": "ClawBio Skill Correctness Bench", "stages": [ "disease-modeling", "target-id", "clinical-development" ], "modalities": [ "cross-modality" ], "task_types": [ "correctness-audit", "safety-audit" ], "description": "Third-party (Biostochastics LLC) benchmark of bio-analysis skills on safety / correctness / honesty. 10 skills \u00d7 182 tests.", "size": { "skills": 10, "tests": 182, "pass_rate_pct": 92.3 }, "primary_paper": { "title": "clawbio_bench README (v0.1.5)", "authors": [ "Biostochastics LLC" ], "year": 2026, "doi": "N/A \u2014 repo", "citations": 5 }, "official_url": "https://clawbio.ai/benchmarks.html", "github_url": "https://github.com/biostochastics/clawbio_bench", "leaderboard_url": "https://clawbio.ai/benchmarks.html", "license": "MIT", "first_release": "2026-04", "last_updated": "2026-05-03", "rubric": { "rigor": 5, "coverage": 2, "maintenance": 5, "adoption": 2, "quality": 4, "accessibility": 5, "industry_relevance": 3 }, "flags": [], "notes": "Independent third-party bench structurally precludes self-reference. Coverage narrow but rigor exemplary.", "related_benchmarks": [], "expert_ids": [], "group_ids": [ "clawbio", "biostochastics" ], "hosted_by": [ "clawbio" ], "composite_score": 74.2 }, { "id": "herg-classifier-bench", "name": "hERG (cardio-tox) TDC", "stages": [ "ind-enabling", "lead-id-admet" ], "modalities": [ "small-molecule" ], "task_types": [ "toxicity-classification" ], "description": "Cardiac tox benchmark (hERG inhibition) \u2014 standardized from Wang et al.", "size": { "molecules": 655, "assays": 1 }, "primary_paper": { "title": "Therapeutics Data Commons", "authors": [ "Huang K", "Fu T", "et al." ], "year": 2021, "doi": "10.48550/arXiv.2102.09548", "citations": 620 }, "official_url": "https://tdcommons.ai/single_pred_tasks/tox/#herg", "github_url": "https://github.com/mims-harvard/TDC", "leaderboard_url": "https://tdcommons.ai/benchmark/admet_group/24herg/", "license": "MIT", "first_release": "2021-02", "last_updated": "2025-01", "rubric": { "rigor": 4, "coverage": 2, "maintenance": 4, "adoption": 4, "quality": 3, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Small but widely benchmarked. Industry pairs with SafetyPanel-5.", "related_benchmarks": [ "tdc-admet", "ames" ], "expert_ids": [ "kexin-huang" ], "group_ids": [ "tdc" ], "hosted_by": [ "tdc", "insilico-scienceaibench", "insilico-ddb" ], "composite_score": 73.9 }, { "id": "dili-ldi", "name": "DILI / LD50 Zhu", "stages": [ "ind-enabling", "lead-id-admet" ], "modalities": [ "small-molecule" ], "task_types": [ "toxicity-regression" ], "description": "Drug-induced liver injury + rat LD50 (Zhu) \u2014 standard acute tox benchmarks.", "size": { "dili_molecules": 475, "ld50_molecules": 7385 }, "primary_paper": { "title": "Quantitative structure-activity relationship modeling of rat acute toxicity", "authors": [ "Zhu H", "Martin TM", "Ye L", "et al." ], "year": 2009, "doi": "10.1021/tx900189p", "citations": 460 }, "official_url": "https://tdcommons.ai/single_pred_tasks/tox/#dili", "github_url": "https://github.com/mims-harvard/TDC", "leaderboard_url": "https://tdcommons.ai/benchmark/admet_group/", "license": "MIT", "first_release": "2021-02", "last_updated": "2024-12", "rubric": { "rigor": 4, "coverage": 2, "maintenance": 4, "adoption": 4, "quality": 3, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Essential IND-enabling endpoints.", "related_benchmarks": [ "tdc-admet", "tox21", "toxcast" ], "expert_ids": [ "kexin-huang" ], "group_ids": [ "tdc" ], "hosted_by": [ "tdc", "insilico-ddb" ], "composite_score": 73.9 }, { "id": "scgpt-bench", "name": "scGPT Evaluation Suite", "stages": [ "virtual-cell" ], "modalities": [ "cross-modality" ], "task_types": [ "cell-type-annotation", "perturbation-prediction" ], "description": "Evaluation shipped with scGPT foundation model (cell type annotation, GRN, perturbation).", "size": { "cells": 33000000, "tasks": 5 }, "primary_paper": { "title": "scGPT: toward building a foundation model for single-cell multi-omics using generative AI", "authors": [ "Cui H", "Wang C", "Maan H", "Pang K", "Luo F", "Wang B" ], "year": 2024, "doi": "10.1038/s41592-024-02201-0", "citations": 520 }, "official_url": "https://github.com/bowang-lab/scGPT", "github_url": "https://github.com/bowang-lab/scGPT", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2023-05", "last_updated": "2025-01", "rubric": { "rigor": 3, "coverage": 3, "maintenance": 4, "adoption": 5, "quality": 3, "accessibility": 5, "industry_relevance": 3 }, "flags": [ "self_referential" ], "notes": "Evaluation dominated by authors' own model \u2014 flagged self-referential. Pair with OpenProblems for fair comparison.", "related_benchmarks": [ "openproblems-perturbation" ], "expert_ids": [ "bo-wang" ], "group_ids": [ "wang-lab-toronto" ], "hosted_by": [], "composite_score": 73.7 }, { "id": "ctod", "name": "CT-Outcome (TrialBench v2)", "stages": [ "phase-2", "phase-3" ], "modalities": [ "cross-modality" ], "task_types": [ "trial-outcome" ], "description": "Updated 2024 trial outcome benchmark with temporal splits.", "size": { "trials": 25000 }, "primary_paper": { "title": "TrialBench: Multi-Modal Artificial Intelligence-Ready Clinical Trial Datasets", "authors": [ "Chen J", "Hu Y", "Wang Y", "et al." ], "year": 2024, "doi": "10.48550/arXiv.2407.00631", "citations": 25 }, "official_url": "https://github.com/ML2Health/ML2ClinicalTrials", "github_url": "https://github.com/ML2Health/ML2ClinicalTrials", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2024-07", "last_updated": "2025-03", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 4, "adoption": 2, "quality": 4, "accessibility": 5, "industry_relevance": 4 }, "flags": [], "notes": "Temporal splits are key improvement.", "related_benchmarks": [ "hint-trialbench" ], "expert_ids": [ "yue-wang", "tianfan-fu" ], "group_ids": [ "ml2health" ], "hosted_by": [ "trialbench" ], "composite_score": 73.4 }, { "id": "dude", "name": "DUD-E", "stages": [ "hit-id" ], "modalities": [ "small-molecule" ], "task_types": [ "virtual-screening" ], "description": "Directory of Useful Decoys \u2014 Enhanced. 102 targets \u00d7 actives + property-matched decoys.", "size": { "targets": 102, "actives": 22886, "decoys": 1411214 }, "primary_paper": { "title": "Directory of Useful Decoys, Enhanced (DUD-E)", "authors": [ "Mysinger MM", "Carchia M", "Irwin JJ", "Shoichet BK" ], "year": 2012, "doi": "10.1021/jm300687e", "citations": 2900 }, "official_url": "http://dude.docking.org/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "CC-BY", "first_release": "2012", "last_updated": "2014", "rubric": { "rigor": 4, "coverage": 4, "maintenance": 1, "adoption": 5, "quality": 3, "accessibility": 5, "industry_relevance": 3 }, "flags": [ "data-leakage-known", "deprecated-recommend-replace" ], "notes": "Well-known analog bias in decoy selection; use LIT-PCBA / PLINDER for fair VS.", "related_benchmarks": [ "lit-pcba", "dekois", "plinder" ], "expert_ids": [ "brian-shoichet", "john-irwin" ], "group_ids": [ "shoichet-lab-ucsf" ], "hosted_by": [ "tdc" ], "composite_score": 72.9 }, { "id": "moses", "name": "MOSES", "stages": [ "lead-id-admet", "developmental-candidate" ], "modalities": [ "small-molecule" ], "task_types": [ "molecule-generation" ], "description": "Molecular Sets \u2014 distribution learning benchmark with 8 reference metrics on ZINC subset.", "size": { "train_set": 1936962, "metrics": 8 }, "primary_paper": { "title": "Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models", "authors": [ "Polykovskiy D", "Zhebrak A", "Sanchez-Lengeling B", "et al." ], "year": 2020, "doi": "10.3389/fphar.2020.565644", "citations": 550 }, "official_url": "https://github.com/molecularsets/moses", "github_url": "https://github.com/molecularsets/moses", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2018-11", "last_updated": "2022-04", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 2, "adoption": 5, "quality": 3, "accessibility": 5, "industry_relevance": 3 }, "flags": [], "notes": "Distribution-learning metrics known to saturate.", "related_benchmarks": [ "guacamol", "pmo" ], "expert_ids": [ "daniil-polykovskiy", "alex-zhavoronkov", "alan-aspuru-guzik" ], "group_ids": [ "insilico-medicine", "matter-lab-toronto" ], "hosted_by": [ "tdc", "moleculenet" ], "composite_score": 72.4 }, { "id": "perturbbench", "name": "PerturbBench", "stages": [ "virtual-cell" ], "modalities": [ "cross-modality" ], "task_types": [ "perturbation-prediction" ], "description": "Benchmark for generalization of perturbation foundation models to unseen genetic perturbations / cell contexts.", "size": { "cells": 400000, "perturbations": 200 }, "primary_paper": { "title": "PerturbBench: Benchmarking Single-Cell Perturbation Foundation Models", "authors": [ "Wu Y", "Barry T", "Wang K", "et al." ], "year": 2024, "doi": "10.48550/arXiv.2412.10091", "citations": 35 }, "official_url": "https://github.com/genentech/PerturbBench", "github_url": "https://github.com/genentech/PerturbBench", "leaderboard_url": "N/A", "license": "Apache-2.0", "first_release": "2024-12", "last_updated": "2025-06", "rubric": { "rigor": 4, "coverage": 3, "maintenance": 3, "adoption": 3, "quality": 4, "accessibility": 4, "industry_relevance": 4 }, "flags": [], "notes": "Pharma-led (Genentech); well-specified eval.", "related_benchmarks": [ "openproblems-perturbation", "scperturb" ], "expert_ids": [ "aviv-regev" ], "group_ids": [ "genentech-gred" ], "hosted_by": [], "composite_score": 71.4 }, { "id": "clintox", "name": "ClinTox", "stages": [ "lead-id-admet", "ind-enabling" ], "modalities": [ "small-molecule" ], "task_types": [ "toxicity-classification" ], "description": "Binary classification of FDA-approved vs. trial-failed-for-toxicity compounds.", "size": { "compounds": 1491 }, "primary_paper": { "title": "Deep learning for drug-induced liver injury", "authors": [ "Xu Y", "Dai Z", "Chen F", "et al." ], "year": 2015, "doi": "10.1021/acs.jcim.5b00238", "citations": 380 }, "official_url": "https://moleculenet.org/datasets-1", "github_url": "https://github.com/deepchem/deepchem", "leaderboard_url": "N/A", "license": "MIT", "first_release": "2015", "last_updated": "2022", "rubric": { "rigor": 3, "coverage": 2, "maintenance": 2, "adoption": 5, "quality": 3, "accessibility": 5, "industry_relevance": 3 }, "flags": [ "data-leakage-known" ], "notes": "Small, binary; saturated. Useful only as sanity check.", "related_benchmarks": [ "tox21", "toxcast" ], "expert_ids": [ "bharath-ramsundar" ], "group_ids": [ "deepchem" ], "hosted_by": [ "moleculenet", "tdc" ], "composite_score": 65.6 }, { "id": "dekois", "name": "DEKOIS 2.0", "stages": [ "hit-id" ], "modalities": [ "small-molecule" ], "task_types": [ "virtual-screening" ], "description": "Decoy sets matched to actives by physicochemical properties for structure-based VS.", "size": { "targets": 81, "actives_per_target": 40 }, "primary_paper": { "title": "DEKOIS 2.0 \u2013 A Public Resource for Benchmarking Structure-based Virtual Screening", "authors": [ "Bauer MR", "Ibrahim TM", "Vogel SM", "Boeckler FM" ], "year": 2013, "doi": "10.1021/ci400115b", "citations": 190 }, "official_url": "http://www.dekois.com/", "github_url": "N/A", "leaderboard_url": "N/A", "license": "CC-BY", "first_release": "2013", "last_updated": "2019", "rubric": { "rigor": 3, "coverage": 3, "maintenance": 2, "adoption": 3, "quality": 3, "accessibility": 4, "industry_relevance": 2 }, "flags": [ "deprecated-recommend-replace" ], "notes": "Historical reference; use LIT-PCBA / PLINDER for modern VS.", "related_benchmarks": [ "dude", "lit-pcba" ], "expert_ids": [ "frank-boeckler" ], "group_ids": [ "tuebingen-boeckler" ], "hosted_by": [], "composite_score": 57.5 } ] }