ProteinGym
TrustAThe best benchmark we have for protein engineering tasks. If your workflow involves designing better variants — which is most of synbio and a growing chunk of drug discovery — ProteinGym performance is more decision-relevant than CASP scores.
What It Measures
Variant effect prediction — how well a model can predict the functional impact of amino acid substitutions. Covers fitness landscapes across hundreds of deep mutational scanning (DMS) datasets.
What It Doesn't Measure
Structure prediction accuracy, binding affinity, or multi-mutant epistatic effects beyond what DMS datasets capture.
Maintainer
Marks Lab (Harvard)
https://proteingym.org/ →Known Limitations
DMS datasets are biased toward well-studied proteins. Fitness assays vary in quality and biological relevance across datasets. Performance on ProteinGym doesn't guarantee performance on your specific protein of interest.