MMPatho: Leveraging Multilevel Consensus and Evolutionary Information for Enhanced Missense Mutation Pathogenic...

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Molecular Genetics

Background:

• Missense mutations (MMs) are crucial for understanding genetic diseases and individual variations.
• Accurate prediction of MM pathogenicity is essential but challenging.

Purpose of the Study:

• To develop a novel computational approach, MMPatho, for enhanced missense mutation pathogenicity prediction.
• To create robust benchmark and blind test datasets for evaluating MM pathogenicity prediction models.

Main Methods:

• Established a large-scale, nonredundant MM benchmark dataset and a focused blind test set.
• Extracted variant-level, amino acid-level, and genome-level features using Ensembl VEP and dbNSFP.
• Utilized protein sequence encoding and extracted embeddings from ESM-1b and ProtTrans-T5 for mutant sites.
• Developed two models, ConsMM (XGBoost with SHAP) and EvoIndMM (incorporating protein language embeddings).

Main Results:

• MMPatho models (ConsMM and EvoIndMM) achieved high performance on a blind test set (AUROC 0.9836-0.9854, AUPR 0.9852-0.9902).
• The models demonstrated superiority in predicting missense mutation pathogenicity.
• A web server was developed for public access to MMPatho prediction and data.

Conclusions:

• MMPatho offers a superior computational approach for predicting missense mutation pathogenicity.
• The developed datasets and web server facilitate further research in genetic disease and variant interpretation.
• Integrating evolutionary information and protein language embeddings significantly enhances predictive capabilities.