Omics and molecular biologyRelease 2026-11-02
Which biological model works best for your task? Published and reproduced results, linked to their sources and organised by biological question.
New. We now flag where studies disagree about the same models, and show what we are doing to resolve each case. See contested results
121 open disagreements between studies, 5 settled in this release.
Biological models and model families. Evaluated versions, configurations and pipelines are linked separately.
Choosing a biological model for a task is harder than it should be. Results are scattered across papers, preprints and benchmark suites. Each uses its own datasets, splits and metrics, so numbers from two papers often cannot be compared, and new models appear faster than anyone can read about them.
This database brings those results together. Every value links to the paper, table or run it came from. Results are organised around use cases, so you can start from the biological question you are trying to answer rather than from a model name. Results measured under different protocols are kept apart instead of being merged into one ranking, and where the evidence for a use case is thin or missing, the page says so. Where published results leave a gap that matters, we run the benchmark ourselves and publish the code.
The goal is a trusted, current reference for how biological models perform on each task: kept up to date as the field moves, open to researchers and their AI agents, and correctable by anyone who brings better evidence.
Published evaluations and rewire evaluations are records in the same database, with their origin shown beside each result. This release includes 49,377 source-checked published results and 12 results from existing rewire runs.
Source checked does not mean independently reproduced. Comparisons require compatible data, protocols and metrics. Missing details stay visible.
Check whether the data supports replaying a metric, investigating an unexpected result or testing an explanation independently.
Readiness and investigations →Read the protocol, coverage and limitations behind our corrected splice-variant evaluation.
MFASS v2 evaluation →MFASS v1 is superseded. Its original tables and methods remain available as an archived report.
Archived MFASS v1 report →86,212 records across 17 record types. These counts describe catalogue coverage, not model performance.
54 of 105 benchmarks have linked evaluations in this release. Zero means no evaluation is linked here, not that the benchmark has never been used.
This is a dated collection, not an exhaustive model census. Models, methods, evaluated configurations and pipelines are listed separately. Counts describe records, not unique checkpoints or independent experiments.
Latest attempt: blocked (1 October 2026). This does not advance the completed sweep date.
A monthly sweep does not mean every record was reverified.
Release 2026-11-02-a7f03c9e55a5 · 2026-11-02
Downloads are hosted on GitHub at the pinned data revision and compressed with gzip.
The original 100-paper collection is retained for citation history. These files include six result rows excluded from the current omics scope; use the database release above for the reviewed collection.