MissenseViewer: On interactive visualization of pathogenicity predictions

Junjie Xu, Stephanie Portelli, Aaron Kovacs, Ashar Malik and David B. Ascher

Abstract: Recent years have seen several advancements in the area of missense mutation effect prediction. Leading these developments are popular deep learning methods, ESM1b and AlphaMissense. Both methods have been employed to predict protein pathogenicity as a result of single-point mutations across all sites in the human proteome. While these results have been made publicly available, their full potential remains underexplored. To address this, we present MissenseViewer, a unified dashboard that consolidates pathogenicity predictions from these programs and now includes predictions from VESPA, a state-of-the-art T5 mutation effect predictor, further enhancing the value of this resource. Using Uniprot accessions, users can visualize substitution sites on both experimentally determined and AlphaFold-predicted protein structures. A susceptibility score is assigned to each site and displayed to assess the likelihood of pathogenicity. Additionally, information from InterPro-Pfam is integrated, allowing users to visualize the domain context of substitution sites. To expand functionality, the app now renders plots highlighting average susceptibility scores for residues within ligand-binding sites for each protein. This feature enables users to explore potential relationships between mutations and ligand-binding, potentially capturing implicit effects on binding dynamics. Altogether, MissenseViewer is a valuable resource for the research community, providing comprehensive and accessible insights into missense mutation effects.