PmmNDD: Predicting the Pathogenicity of Missense Mutations in Neurodegenerative Diseases via Ensemble Learning

Accurately distinguishing between pathogenic and benign mutations continues to pose a significant challenge in the clinical genetic testing of patients with neurodegenerative diseases (NDDs). In theory, computational methods have the potential to facilitate the interpretation of genetic variants in NDDs on a large scale. However, individual tools often exhibit disagreements, biases, and variations in quality. As a result, the predictions derived from them are considered insufficiently reliable. In this study, we developed PmmNDD, an ensemble method for predicting pathogenicity of missense variants in NDDs. PmmNDD integrated the prediction scores from other methods along with amino acid characteristics as features, and was constructed with the categorical boosting (CatBoost) model. The stability and generalization ability of PmmNDD were validated through leave-one-gene-out cross-validation and independent test. We also demonstrated Pmm-NDD's superior performance over 20 other methods. Furthermore, we provided pre-computed PmmNDD scores for all possible NDDs missense variants to facilitate the identification of pathogenic variants in the sea of rare variants discovered as sequencing studies expand in scale. In summary, our work suggests that models from ensemble learning can provide valuable independent evidence for NDD mutation interpretation that will be widely useful in research and clinical scenarios.