MissenseHMM: state-based annotations for missense variants through joint modeling of pathogenicity scores

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Numerous computational tools exist for predicting missense variant pathogenicity.
• Integrating information from diverse predictors is crucial for accurate variant classification.

Purpose of the Study:

• To develop MissenseHMM, a novel method that learns combinatorial patterns from multiple pathogenicity predictors.
• To annotate a large set of missense variants using MissenseHMM for improved interpretation.

Main Methods:

• Applied MissenseHMM to 43 existing pathogenicity predictors.
• Trained MissenseHMM to identify 20 distinct states based on predictor score patterns.
• Annotated over 70 million missense variants with the learned states.

Main Results:

• MissenseHMM states revealed distinct patterns in predictor scores, amino acid substitutions, and genomic annotations.
• Annotations derived from MissenseHMM improved associations with clinical pathogenic variants and deep mutational scanning data.
• The study provided insights into the performance of various protein language models.

Conclusions:

• MissenseHMM serves as a valuable annotation resource for missense variant interpretation.
• This approach enhances the utility of individual pathogenicity predictors.
• The learned states offer a nuanced understanding of variant pathogenicity.