Definition
Plain language
A benchmark for predicting how changing a protein's building blocks affects how it works.
As stated in the literature
A benchmark of hundreds of deep mutational scanning datasets evaluating models on predicting the functional effects of protein mutations, scored by rank correlation with experiment.
Why it matters: It's a shared yardstick for mutation-effect prediction, helping researchers tell whether a new method genuinely improves on the last.
For example, it checks how well a model predicts which mutations make a protein work better or worse, comparing its rankings to lab measurements.
Heard on the show
“… ProteinGym is a benchmark for predicting how mutations to a protein affect its function — two-hundred-seventeen …”Episode 095 — Seven Wins to Zero: How Organizing AI Agents Like a Lab Changes the Search