Variant pathogenicity prediction based on the ESGMM algorithm

• DOI: 10.1007/s10791-024-09487-w
• Publication Date: 2024-12-09T16:45:07Z
• Authors: Jing Liu; Chen Zhang; Xiyi Liu; Yu Zhang; Jin Liu; Xiao Guan

Modeling the functional impact of sequence variation is a critical issue for both understanding and developing proteins. An Evolutionary Sequence Gaussian Mixture Model (ESGMM) predicting variant pathogenicity presented in this paper. The model trained on 2715 clinical proteins their homologous sequences, using Transformer-based protein language to discover evolutionary patterns amino acids from multiple alignment (MSA). To fully mine deep information MSA two-dimensional data, an axial attention mechanism introduced during training. estimates probability all variants compared wild type calculates scores. categorize variations as pathogenic or benign, global–local mixture then constructed each variant, ESGMM scores are produced employing combination global local information. Particle swarm optimization (PSO) optimize further quantify uncertainty classification, which enhances prediction precision. Experimental results demonstrate superiority optimized algorithm variants.