{"success":true,"data":{"author":"Genentech (Nair, Gunsalus, Orcutt-Jahns, Rossen et al.)","bucketAccessType":"public","bucketName":"Genentech/compbiobench-data-v1","bucketPrefix":"","category":"Genomics & Bioinformatics","cloudProvider":"huggingface","createdAt":"2026-11-17T14:12:21.591Z","description":"CompBioBench v1 from Genentech is a benchmark of 100 tasks for evaluating agentic AI systems in computational biology. To make noisy biological analysis objectively gradable, tasks are built from synthetic or augmented data and from real datasets whose metadata has been scrambled or scrubbed, so each problem has a single ground-truth answer that still needs multi-step reasoning, tool use, bespoke code and queries to external resources. Tasks span genomics, transcriptomics, epigenomics, single-cell analysis, human genetics and machine learning workflows, and were curated by domain experts across a range of difficulty. The repository holds compbiobench.v1.tsv (questions and metadata) and data/ with the input files each task needs, about 12 GB in total; answers are scored on the public leaderboard Space (Genentech/compbiobench-leaderboard-v1). The same record is on Zenodo (10.5281/zenodo.19443186), where the authors release it under CC BY 4.0; cite Nair, Gunsalus, Orcutt-Jahns, Rossen et al. (2026).","displayTitle":"CompBioBench v1: 100 verifiable computational biology tasks for AI agents, with input data","doi":"10.5281/zenodo.19443186","huggingFaceUrl":"https://huggingface.co/datasets/Genentech/compbiobench-data-v1","id":"3cfc75a4-b130-5b4c-bf71-a0597f990164","license":"CC-BY-4.0","licenseEvidenceUrl":"https://zenodo.org/records/19443186","objectCountEstimate":100,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"Genentech/compbiobench-data-v1\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"Genentech/compbiobench-data-v1\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=Genentech, license:cc-by-4.0, size_categories:n<1K, downloads=1047"],"format":"mixed","generatedAt":"2026-11-17T14:23:04.439Z","layout":{"example":"compbiobench.v1.tsv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"question_id\tcurator_name\tdomain\tquestion_style\tskills_tested\tquestion\tinternet_required\tgpu_preferred\tfile_paths\nensembl-grab-q1\tLG\tGenomics\tRetrieval\tAPI/Web Fetching\tConvert all of the following to Ensembl gene IDs. Provide the ID without the Ensembl version (e.g., ENSGXXXXXXXXXXXX not ENSGXXXXXXXXXXXX.XX). Respond with semi-colon separated values, no spaces. Inputs: 1. NM_001276266.2 2. TERB2 3. ENST00000267814 4. chr15:45167214-45187966 (hg38) 5. GeneID:9153 6. NP_922946.1 7. MIM:617658 8. BEN domain co","encoding":"utf8","key":"compbiobench.v1.tsv","truncatedAt":512}],"sampleObjects":["compbiobench.v1.tsv","data/10x_pbmc68k_reduced.h5ad","data/E055_15_coreMarks_dense.bed.gz","data/ENCFF333TAT.bed.gz","data/ENCFF356LFX.bed.gz"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\npretty_name: compbiobench-v1\ntags:\n- biology\n- compbio\n- genomics\n- benchmark\n- agent\n- bioinformatics\nsize_categories:\n- n<1K\n---\n\n# CompBioBench v1: A benchmark of 100 diverse, verifiable questions for agents for computational biology\n\nWe introduce CompBioBench v1, a benchmark of 100 diverse tasks for evaluating agentic systems in computational biology. Unlike mathematics and programming, which more readily admit systematic verification, biological data are inherently noisy and open to interpretation. To enable objective evaluation without reducing tasks to prescriptive checklists, we propose a new benchmark-construction strategy based on synthetic/augmented data and metadata scrambling/scrubbing of real datasets to create challenging problems with a single ground-truth answer that require multi-step reasoning, tool use, bespoke code, and interaction with real-world external resources. The benchmark spans genomics, transcriptomics, epigenomics, single-cell analysis, human genetics, and machine learning workflows. Questions are curated by domain experts to cover a broad range of skills with varying difficulty.\n\nThis record contains all questions, metadata, and input data files associated with CompBioBench v1. Also available on [Zenodo](https://zenodo.org/records/19443186). You can submit your answers for evaluation at this [Leaderboard](https://huggingface.co/spaces/Genentech/compbiobench-leaderboard-v1).","encoding":"utf8","key":"README.md","name":"README.md","size":1452,"truncated":false}],"sourceObjectKey":"compbiobench.v1.tsv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/Genentech/compbiobench-data-v1/compbiobench.v1.tsv\", sep='\\t')\ndf.head()","starterSamplePath":"compbiobench.v1.tsv","totalObjects":100},"summary":"CompBioBench v1 from Genentech is a benchmark of 100 tasks for evaluating agentic AI systems in computational biology.","title":"CompBioBench v1: 100 verifiable computational biology tasks for AI agents, with input data","updatedAt":"2026-11-17T20:19:07.920Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/Genentech/compbiobench-data-v1"}}