Extended Data Fig. 4: Evo 2 enables accurate human clinical variant effect prediction. | Nature

Extended Data Fig. 4: Evo 2 enables accurate human clinical variant effect prediction.

From: Genome modelling and design across all domains of life with Evo 2

Extended Data Fig. 4: Evo 2 enables accurate human clinical variant effect prediction.

(a) ClinVar variants stratified by PhyloP score and variant type (coding or noncoding, SNV or non-SNV). (b) ClinVar variants stratified by proximity to splice sites, where “near” is defined as within 5 bp or 3 bp to donor or acceptor sites, respectively, and “far” is defined as otherwise. Results are also separated by variant type (coding or noncoding, SNV or non-SNV). Because there were only 22 coding non-SNVs that are near splice sites and all of these variants have the same label (“pathogenic”), we excluded this condition in our evaluations due to the limited sample size and the inapplicability of binary classification metrics like AUROC and AUPRC. (c) Zero-shot noncoding ClinVar evaluations after filtering out all variants with either a SpliceAI score ≥ 0.1 or a Pangolin score ≥ 0.1 (i.e., keeping variants with low SpliceAI and low Pangolin scores).

Back to article page