Prediction of recessive inheritance for missense variants in human disease

Abstract

The prediction of pathogenic human missense variants has improved in recent years, but a more granular level of variant characterization is required. Further axes of information need to be incorporated in order to advance the genotype-to-phenotype map. Recent efforts have developed mode of inheritance prediction tools; however, these lack robust validation and their discrimination performance does not support clinical utility, with evidence of them being fundamentally insensitive to recessive acting diseases. Here, we present MOI-Pred, a three-way variant-level mode of inheritance prediction tool aimed at recessive identification for missense variants. MOI-Pred shows strong ability to discriminate missense variants causing autosomal recessive disease (area under the receiver operating characteristic (AUROC)=0.99 and sensitivity=0.85) in an external validation set. Additionally, we introduce an electronic health record (EHR)-based validation approach using real-world clinical data and show that our recessive predictions are enriched for recessive associations with human diseases, demonstrating utility of our method. Mode of inheritance predictions - pathogenic for autosomal recessive (AR) disease, pathogenic for autosomal dominant (AD) disease, or benign - for all possible missense variants in the human genome are available at https://github.com/rondolab/MOI-Pred/.