Utility of gene-specific algorithms for predicting pathogenicity of uncertain gene variants

Abstract

The rapid advance of gene sequencing technologies has produced an unprecedented rate of discovery of genome variation in humans. A growing number of authoritative clinical repositories archive gene variants and disease phenotypes, yet there are currently many more gene variants that lack clear annotation or disease association. To date, there has been very limited coverage of genespecific predictors in the literature. Here the evaluation is presented of "gene-specific" predictor models based on a nai{dotless} ̈ve Bayesian classifier for 20 geneedisease datasets, containing 3986 variants with clinically characterized patient conditions. The utility of genespecific prediction is then compared with "all-gene" generalized prediction and also with existing popular predictors. Gene-specific computational prediction models derived from clinically curated gene variant disease datasets often outperform established generalized algorithms for novel and uncertain gene variants.