Loading dataset details…
harbor run -d nextflow-frontier/bioml-benchharbor run -d nextflow-frontier/bioml-bench22 biomedical machine-learning tasks from BioML-bench, packaged for local Harbor runs.
harbor run -d nextflow-frontier/bioml-bench22 biomedical machine-learning tasks from BioML-bench, packaged for local Harbor runs.
harbor run -d nextflow-frontier/bioml-benchPaper · Original benchmark · Nextflow
BioML-bench evaluates AI agents on end-to-end biomedical machine learning. This Nextflow collection packages 22 tasks for Harbor across drug discovery, protein engineering, single-cell analysis and medical imaging.
These tasks require local Docker, your own lawfully held pinned data packets, and the included grading adapter. Data and protected answers are not distributed. Hosted “New Job” is not supported.
uv tool install harbor==0.25.0
harbor download nextflow-frontier/bioml-bench@1.0.0 -o ./bioml-benchFollow local-runner.md in the download to configure the packets and run with bioml_grader_lifecycle.py. Stock Harbor agents and resumed trials are not supported by this grading integration.
The Per-task results board contains 44 historical results from Rosalind and Claude Science. These runs predate the packaged tasks. Protocols and resources differ, so results are reported per task without an overall ranking. See per-task-results.md and each run's provenance for details.
Miller HE, Greenig M, Tenmann B, Wang B (2025). BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML. bioRxiv. doi:10.1101/2025.09.01.673319.
Paper · Original benchmark · Nextflow
BioML-bench evaluates AI agents on end-to-end biomedical machine learning. This Nextflow collection packages 22 tasks for Harbor across drug discovery, protein engineering, single-cell analysis and medical imaging.
These tasks require local Docker, your own lawfully held pinned data packets, and the included grading adapter. Data and protected answers are not distributed. Hosted “New Job” is not supported.
uv tool install harbor==0.25.0
harbor download nextflow-frontier/bioml-bench@1.0.0 -o ./bioml-benchFollow local-runner.md in the download to configure the packets and run with bioml_grader_lifecycle.py. Stock Harbor agents and resumed trials are not supported by this grading integration.
The Per-task results board contains 44 historical results from Rosalind and Claude Science. These runs predate the packaged tasks. Protocols and resources differ, so results are reported per task without an overall ranking. See per-task-results.md and each run's provenance for details.
Miller HE, Greenig M, Tenmann B, Wang B (2025). BioML-bench: Evaluation of AI Agents for End-to-End Biomedical ML. bioRxiv. doi:10.1101/2025.09.01.673319.