The predictions and interpretations from this tool are computational hypotheses generated by machine learning models.They should not be treated as clinical diagnoses or used as the sole basis for medical decisions.
Each point is one missense variant. Coordinates come from running ESM-C (6B) on the mutant protein sequence and reading the layer-78 embedding at the mutated residue, then projecting those high-dimensional vectors to 2D with UMAP. Variants that land near each other have similar ESM-C representations.
Select a colormap from the menu on the bottom left of the site. The default colormap is the model's predicted pathogenicity — the ESM-C covariance probe's P(pathogenic) for that variant, read at layer 80. The clinical class colormap labels known pathogenic or benign variants, where ClinVar has them annotated.
The mutant residue colormap describes the substitution itself: the amino acid the variant introduces. Colors are grouped by side-chain chemistry.