BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks
Each agent runs four independent replicates on each of the 9 drug discovery tasks. Each submission is graded with the task's original metric and mapped to that task's public leaderboard percentile; Table 4 reports domain means and rates.
Overview
An explanatory profile has not yet been reviewed for this record.
Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.
10 recorded evaluations, 34 metric rows. A comparison chart has not yet been validated for these results. The table retains the individual findings and their sources.
Results
Results are available, but no reviewed comparison panel is linked in this release.
All evaluations
10 evaluations · 34 results. Different protocols are not a single leaderboard.
Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks · Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks)
Sorted by Leaderboard percentile (mean over tasks in the domain; non-penalized: runs without a valid submission omitted) (higher is better). The best value in each column is highlighted. Decimals are rounded for display; each value links to the printed value and its source.
| Tested configuration | Leaderboard percentile (mean over tasks in the domain; non-penalized: runs without a valid submission omitted) | Leaderboard percentile (mean over tasks in the domain; penalized: runs without a valid submission scored as percentile 0) | Proportion (share of tasks where the agent reaches a public leaderboard medal threshold, MLE-bench definition (Section 3.4); non-penalized: runs without a valid submission omitted) | Proportion (share of tasks where the agent reaches a public leaderboard medal threshold, MLE-bench definition (Section 3.4); penalized: runs without a valid submission scored as percentile 0) | Proportion (share of tasks where the agent's mean score exceeds the median human leaderboard submission (Section 3.4 definition); non-penalized: runs without a valid submission omitted) | Proportion (share of tasks where the agent's mean score exceeds the median human leaderboard submission (Section 3.4 definition); penalized: runs without a valid submission scored as percentile 0) | Proportion (share of tasks that yield a valid, gradable submission (Section 3.4 definition); penalized: runs without a valid submission scored as percentile 0) |
|---|---|---|---|---|---|---|---|
| Biomni v0.0.5, Claude Sonnet 4 backend (BioML-bench)Evaluation | 32.55 ± 8.61% (best in column) | Not reported | 13.9 ± 11.1% (best in column) | Not reported | 30.6 ± 12.3% (best in column) | Not reported | Not reported |
| AIDE v6.3.3, GPT-4.1 backend (BioML-bench)Evaluation | 31.06 ± 8.88% | Not reported | 7.4 ± 7.4% | Not reported | 30.6 ± 13.3% (best in column) | Not reported | Not reported |
| STELLA, OpenRouter default backend (BioML-bench)Evaluation | 28.84 ± 7.84% | Not reported | 13.9 ± 7.3% (best in column) | Not reported | 25.0 ± 11.0% | Not reported | Not reported |
| MLAgentBench, GPT-4.1 backend (BioML-bench)Evaluation | 22.45 ± 5.93% | Not reported | 5.6 ± 3.7% | Not reported | 16.7 ± 9.3% | Not reported | Not reported |
| Dummy agent (BioML-bench)Evaluation | 1.11 ± 1.11% | Not reported | 0.0 ± 0.0% | Not reported | 0.0 ± 0.0% | Not reported | Not reported |
| AIDE v6.3.3, GPT-4.1 backend (BioML-bench)Evaluation | Not reported | 24.75 ± 7.23% | Not reported | 5.6 ± 5.6% | Not reported | 25.0 ± 11.0% | 80.6% |
| Biomni v0.0.5, Claude Sonnet 4 backend (BioML-bench)Evaluation | Not reported | 32.55 ± 8.61% (best in column) | Not reported | 13.9 ± 11.1% (best in column) | Not reported | 30.6 ± 12.3% (best in column) | 100% (best in column) |
| Dummy agent (BioML-bench)Evaluation | Not reported | 1.11 ± 1.11% | Not reported | 0.0 ± 0.0% | Not reported | 0.0 ± 0.0% | Not reported |
| MLAgentBench, GPT-4.1 backend (BioML-bench)Evaluation | Not reported | 22.45 ± 5.93% | Not reported | 5.6 ± 3.7% | Not reported | 16.7 ± 9.3% | 100% (best in column) |
| STELLA, OpenRouter default backend (BioML-bench)Evaluation | Not reported | 28.84 ± 7.84% | Not reported | 13.9 ± 7.3% (best in column) | Not reported | 25.0 ± 11.0% | 100% (best in column) |
All 34 result rows with coverage, uncertainty and sources
Filter evaluations
Applied filters: All linked evaluations
| Tested configuration | Protocol and dataset | Finding | Evidence and details |
|---|---|---|---|
| Configuration: AIDE v6.3.3, GPT-4.1 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 30.6 ± 13.3% Proportion (share of tasks where the agent's mean score exceeds the median human leaderboard submission (Section 3.4 definition); non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 13.3 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceAIDE on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'AIDE', column 'Above Median (%)' |
| Configuration: AIDE v6.3.3, GPT-4.1 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 7.4 ± 7.4% Proportion (share of tasks where the agent reaches a public leaderboard medal threshold, MLE-bench definition (Section 3.4); non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 7.4 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceAIDE on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'AIDE', column 'Any Medal (%)' |
| Configuration: AIDE v6.3.3, GPT-4.1 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 31.06 ± 8.88% Leaderboard percentile (mean over tasks in the domain; non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 8.88 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceAIDE on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'AIDE', column 'Leaderboard Percentile' |
| Configuration: Biomni v0.0.5, Claude Sonnet 4 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 30.6 ± 12.3% Proportion (share of tasks where the agent's mean score exceeds the median human leaderboard submission (Section 3.4 definition); non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 12.3 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceBiomni on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'Biomni', column 'Above Median (%)' |
| Configuration: Biomni v0.0.5, Claude Sonnet 4 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 13.9 ± 11.1% Proportion (share of tasks where the agent reaches a public leaderboard medal threshold, MLE-bench definition (Section 3.4); non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 11.1 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceBiomni on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'Biomni', column 'Any Medal (%)' |
| Configuration: Biomni v0.0.5, Claude Sonnet 4 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 32.55 ± 8.61% Leaderboard percentile (mean over tasks in the domain; non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 8.61 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceBiomni on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'Biomni', column 'Leaderboard Percentile' |
| Configuration: Dummy agent (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 0.0 ± 0.0% Proportion (share of tasks where the agent's mean score exceeds the median human leaderboard submission (Section 3.4 definition); non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 0.0 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceDummy on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'Dummy', column 'Above Median (%)' |
| Configuration: Dummy agent (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 0.0 ± 0.0% Proportion (share of tasks where the agent reaches a public leaderboard medal threshold, MLE-bench definition (Section 3.4); non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 0.0 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceDummy on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'Dummy', column 'Any Medal (%)' |
| Configuration: Dummy agent (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 1.11 ± 1.11% Leaderboard percentile (mean over tasks in the domain; non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 1.11 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceDummy on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'Dummy', column 'Leaderboard Percentile' |
| Configuration: MLAgentBench, GPT-4.1 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 16.7 ± 9.3% Proportion (share of tasks where the agent's mean score exceeds the median human leaderboard submission (Section 3.4 definition); non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 9.3 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceMLAgentBench on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'MLAgentBench', column 'Above Median (%)' |
| Configuration: MLAgentBench, GPT-4.1 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 5.6 ± 3.7% Proportion (share of tasks where the agent reaches a public leaderboard medal threshold, MLE-bench definition (Section 3.4); non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 3.7 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceMLAgentBench on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'MLAgentBench', column 'Any Medal (%)' |
| Configuration: MLAgentBench, GPT-4.1 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 22.45 ± 5.93% Leaderboard percentile (mean over tasks in the domain; non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 5.93 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceMLAgentBench on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'MLAgentBench', column 'Leaderboard Percentile' |
| Configuration: STELLA, OpenRouter default backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 25.0 ± 11.0% Proportion (share of tasks where the agent's mean score exceeds the median human leaderboard submission (Section 3.4 definition); non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 11.0 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceSTELLA on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'STELLA', column 'Above Median (%)' |
| Configuration: STELLA, OpenRouter default backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 13.9 ± 7.3% Proportion (share of tasks where the agent reaches a public leaderboard medal threshold, MLE-bench definition (Section 3.4); non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 7.3 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceSTELLA on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'STELLA', column 'Any Medal (%)' |
| Configuration: STELLA, OpenRouter default backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 28.84 ± 7.84% Leaderboard percentile (mean over tasks in the domain; non-penalized: runs without a valid submission omitted) percent · higher Uncertainty: ± 7.84 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceSTELLA on BioML-bench drug discovery (non-penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Non-penalized', domain 'Drug Discovery', row 'STELLA', column 'Leaderboard Percentile' |
| Configuration: AIDE v6.3.3, GPT-4.1 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 25.0 ± 11.0% Proportion (share of tasks where the agent's mean score exceeds the median human leaderboard submission (Section 3.4 definition); penalized: runs without a valid submission scored as percentile 0) percent · higher Uncertainty: ± 11.0 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceAIDE on BioML-bench drug discovery (penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Penalized', domain 'Drug Discovery', row 'AIDE', column 'Above Median (%)' |
| Configuration: AIDE v6.3.3, GPT-4.1 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 5.6 ± 5.6% Proportion (share of tasks where the agent reaches a public leaderboard medal threshold, MLE-bench definition (Section 3.4); penalized: runs without a valid submission scored as percentile 0) percent · higher Uncertainty: ± 5.6 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceAIDE on BioML-bench drug discovery (penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Penalized', domain 'Drug Discovery', row 'AIDE', column 'Any Medal (%)' |
| Configuration: AIDE v6.3.3, GPT-4.1 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 80.6% Proportion (share of tasks that yield a valid, gradable submission (Section 3.4 definition); penalized: runs without a valid submission scored as percentile 0) percent · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceAIDE on BioML-bench drug discovery (penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Penalized', domain 'Drug Discovery', row 'AIDE', column 'Completion Rate (%)' |
| Configuration: AIDE v6.3.3, GPT-4.1 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 24.75 ± 7.23% Leaderboard percentile (mean over tasks in the domain; penalized: runs without a valid submission scored as percentile 0) percent · higher Uncertainty: ± 7.23 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceAIDE on BioML-bench drug discovery (penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Penalized', domain 'Drug Discovery', row 'AIDE', column 'Leaderboard Percentile' |
| Configuration: Biomni v0.0.5, Claude Sonnet 4 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 30.6 ± 12.3% Proportion (share of tasks where the agent's mean score exceeds the median human leaderboard submission (Section 3.4 definition); penalized: runs without a valid submission scored as percentile 0) percent · higher Uncertainty: ± 12.3 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceBiomni on BioML-bench drug discovery (penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Penalized', domain 'Drug Discovery', row 'Biomni', column 'Above Median (%)' |
| Configuration: Biomni v0.0.5, Claude Sonnet 4 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 13.9 ± 11.1% Proportion (share of tasks where the agent reaches a public leaderboard medal threshold, MLE-bench definition (Section 3.4); penalized: runs without a valid submission scored as percentile 0) percent · higher Uncertainty: ± 11.1 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceBiomni on BioML-bench drug discovery (penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Penalized', domain 'Drug Discovery', row 'Biomni', column 'Any Medal (%)' |
| Configuration: Biomni v0.0.5, Claude Sonnet 4 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 100% Proportion (share of tasks that yield a valid, gradable submission (Section 3.4 definition); penalized: runs without a valid submission scored as percentile 0) percent · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceBiomni on BioML-bench drug discovery (penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Penalized', domain 'Drug Discovery', row 'Biomni', column 'Completion Rate (%)' |
| Configuration: Biomni v0.0.5, Claude Sonnet 4 backend (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 32.55 ± 8.61% Leaderboard percentile (mean over tasks in the domain; penalized: runs without a valid submission scored as percentile 0) percent · higher Uncertainty: ± 8.61 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceBiomni on BioML-bench drug discovery (penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Penalized', domain 'Drug Discovery', row 'Biomni', column 'Leaderboard Percentile' |
| Configuration: Dummy agent (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 0.0 ± 0.0% Proportion (share of tasks where the agent's mean score exceeds the median human leaderboard submission (Section 3.4 definition); penalized: runs without a valid submission scored as percentile 0) percent · higher Uncertainty: ± 0.0 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceDummy on BioML-bench drug discovery (penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Penalized', domain 'Drug Discovery', row 'Dummy', column 'Above Median (%)' |
| Configuration: Dummy agent (BioML-bench) | Protocol: BioML-bench drug discovery domain: agent leaderboard percentile over 9 tasks Dataset subset: BioML-bench v0.1-alpha drug discovery tasks (9 tasks) | 0.0 ± 0.0% Proportion (share of tasks where the agent reaches a public leaderboard medal threshold, MLE-bench definition (Section 3.4); penalized: runs without a valid submission scored as percentile 0) percent · higher Uncertainty: ± 0.0 (type not stated). Table 4 caption does not name the spread. Figures 2 and 4 show per-domain leaderboard percentiles as mean ± SEM across tasks; the Table 2 caption refers to the reported SEM. Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceDummy on BioML-bench drug discovery (penalized) lit-agents-20261013-biomlbench-protocol-drug-discovery Aggregation: Not reported BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Table 4 (page 12), block 'Penalized', domain 'Drug Discovery', row 'Dummy', column 'Any Medal (%)' |
Source checking is not independent reproduction. Release 2026-11-02-a7f03c9e55a5.
Methods and evaluation design
Procedure, tasks and evaluated configurations
Recorded evaluations
Each evaluation records what was tested and under which conditions.
- AIDE on BioML-bench drug discovery (non-penalized)
- Biomni on BioML-bench drug discovery (non-penalized)
- Dummy on BioML-bench drug discovery (non-penalized)
- MLAgentBench on BioML-bench drug discovery (non-penalized)
- STELLA on BioML-bench drug discovery (non-penalized)
- AIDE on BioML-bench drug discovery (penalized)
- Biomni on BioML-bench drug discovery (penalized)
- Dummy on BioML-bench drug discovery (penalized)
- MLAgentBench on BioML-bench drug discovery (penalized)
- STELLA on BioML-bench drug discovery (penalized)
Baseline coverage
Reference methods help show what a model adds beyond simple controls. We track a null control and a conventional method for each protocol.
0 of 2 active baseline roles have published Rewire measurements in this release. Measurements on a selected protocol do not establish coverage of an entire suite.
No execution recipe linked to this protocol. Recipe availability does not establish a completed evaluation.
- External evaluations
- 10
Literature evidence is not a Rewire measurement. Executed but unpublished runs and private review status are not included.
Null control
Proposed control: requires review
Select a task-valid null control after reviewing inputs and metric
Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.
This is a suggested selection rule, not a validated method or a measured score.
Conventional reference
Proposed control: requires review
Select an upstream conventional reference after reviewing the full protocol
Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.
This is a suggested selection rule, not a validated method or a measured score.
Protocol coverage CSV (gzip) · Model evaluation matrix (gzip) · Source table (gzip) · Release and checksums (gzip)
Coverage is derived from release 2026-11-02-a7f03c9e55a5. Source citations describe the original records; they do not validate an unreviewed baseline proposal. No results have been generated by this audit.
Run instructions
No runnable recipe has been reviewed for this protocol. Dataset access, model requirements, licences and compute requirements must be checked against its sources before execution.
Evidence
Source checking verifies the cited claim or transcription. It does not establish independent reproduction.
Evidence table
Inspect claims, sources and review details
Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.
One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.
0 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|
No evidence rows match these filters. Choose another scope or clear the search.
Sources and history
Release 2026-11-02-a7f03c9e55a5 · Record review: source checked
1 source record and release history
- BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML · Original source · bioRxiv 2025.09.01.673319 version 3, posted 2025-09-28 (versions 1 and 2 posted 2025-09-04 and 2025-09-07)
Technical metadata and extraction receipts
Stable ID: lit-agents-20261013-biomlbench-protocol-drug-discovery
- areas
- molecular-interactions
- contexts
- research
- protocol
- Task capsule per task: task description, training data, test features and a sample submission. The agent designs, implements and runs ML models in a Docker container with a pre-built conda environment and submits test-set predictions, which are validated and graded with the task's original metric. Four independent replicates per agent and task. Scored in two settings: penalized (runs without a valid submission get percentile 0) and non-penalized (such runs are omitted).
- version
- BioML-bench v0.1-alpha as reported in bioRxiv v3 (Sections 3.1-3.4, Table 4)
- metric
- leaderboard-percentile
- metric direction
- higher
- unit
- percent
- unit detail
- percentile of public leaderboard entries, 0 to 100
- metric definition
- Leaderboard percentile: agent score mapped to the public leaderboard percentile for the task (higher is better). Mean rank: average relative rank of an agent compared to the other agents across tasks, based on leaderboard percentile (lower is better). Above-median: proportion of tasks where the agent's mean score exceeds the median human submission. Any-medal: proportion of tasks where the agent achieves a public leaderboard medal threshold, following the MLE-bench definition. Completion rate: proportion of tasks that yield a valid, gradable submission. (Section 3.4, quoted closely.)
- selection
- Tasks: polaris-pkis2-egfr-wt-c-1, polaris-adme-fang-hclint-1, polaris-adme-fang-hppb-1, polaris-adme-fang-solu-1, tdcommons-cyp2d6-substrate, tdcommons-lipophilicity, tdcommons-herg, tdcommons-bbb-martins, tdcommons-caco2-wang (Table 3).
- replicates
- Four independent replicates per agent-task combination
- hyperparameters
- Compute budget: CPU-only machines with 16 vCPUs and 64 GB RAM and an 8-hour wall-clock limit for non-imaging tasks (Section 3.4)
- source locator
- Sections 3.1, 3.4 and 4.2; Table 3; Table 4 block 'Drug Discovery'
- limitations
- Leaderboard percentile depends on the composition of each task's public leaderboard; the paper describes the leaderboards as largely expert-populated but does not print their size or snapshot dates.; Results are domain means over a small number of tasks; per-task values are only shown in figures.; The plus-or-minus spread in Table 4 is not identified in its caption; Figures 2 and 4 describe the same quantities as mean ± SEM across tasks.; Above Median, Any Medal and Completion Rate values fit counting replicate runs rather than tasks; the denominator is unresolved.; The agents were run by the benchmark's authors, not by their developers.
Related records
- part of: BioML-bench
- uses data: BioML-bench v0.1-alpha drug discovery tasks (9 tasks)
- assessment: AIDE on BioML-bench drug discovery (non-penalized)
- assessment: Biomni on BioML-bench drug discovery (non-penalized)
- assessment: Dummy on BioML-bench drug discovery (non-penalized)
- assessment: MLAgentBench on BioML-bench drug discovery (non-penalized)
- assessment: STELLA on BioML-bench drug discovery (non-penalized)
- assessment: AIDE on BioML-bench drug discovery (penalized)
- assessment: Biomni on BioML-bench drug discovery (penalized)
- assessment: Dummy on BioML-bench drug discovery (penalized)
- assessment: MLAgentBench on BioML-bench drug discovery (penalized)
- assessment: STELLA on BioML-bench drug discovery (penalized)
- parent: BioML-bench task polaris-adme-fang-hclint-1
- parent: BioML-bench task polaris-adme-fang-hppb-1
- parent: BioML-bench task polaris-adme-fang-solu-1
- parent: BioML-bench task polaris-pkis2-egfr-wt-c-1
- parent: BioML-bench task tdcommons-bbb-martins
- parent: BioML-bench task tdcommons-caco2-wang
- parent: BioML-bench task tdcommons-cyp2d6-substrate
- parent: BioML-bench task tdcommons-herg
- parent: BioML-bench task tdcommons-lipophilicity