PPSNV: A Novel Predictor for Pathogenicity of Nonsynonymous SNV based on Ensemble Learning

Abstract

With the next-generation sequencing (NGS) technologies developing, numerous genetic data are available for researchers and clinical doctors. Nonsynonymous single nucleotide variant (nonsynonymous SNV) is a common type of genetic mutation which possibly leads to diseases. However, classifying observed SNVs to benign or pathogenic variants with high confidence remains challenging. Inspired by ensemble learning and Gradient Boosting Decision Tree (GBDT), a machine learning algorithm, we proposed a novel prediction model named PPSNV to identify the pathogenicity of nonsynonymous SNVs. We integrated 14 features to train our model and tested it in two independent datasets. The results showed outstanding performance was achieved by the proposed predictors compared with four commonly used prediction tools.