Prediction of Human Pathogenic Start Loss Variants Based on Multi-channel Features

Start loss variants occur at the start codon that can disrupt the normal translation initiation process, potentially resulting in the production of abnormal protein isoforms. Although numerous computational methods have been developed to aid in the large-scale interpretation of genetic variants, they often show limited predictive accuracy for start-loss variants. A significant limitation of the majority of these methods is their dependence on manually curated features, which restricts their ability to predict variants that have not been studied and characterized. Here, we introduce StartPred, a novel prediction method specifically designed to identify pathogenic start loss variants. It effectively integrated multichannel features from both reference and mutated sequences to facilitate a thorough characterization of the functional implications of start loss variants. Our experimental results show that StartPred outperforms 13 other methods in predicting the pathogenicity of start loss variants. It also excels at identifying the pathogenicity of variants occurring in genes that have not been extensively studied. In addition, our results indicate that certain start loss variants may be closely associated with neurodegenerative diseases, highlighting the potential of StartPred to identify risk loci. This work lays a foundation for accurately deciphering the functional impact of start loss variants in the human genome. StartPred is available at https://github.com/xialab-ahu/StartPred.