PMut

PMut predicts the pathological character of single point amino acid mutations to support interpretation of disease-associated and Mendelian protein variants.

Key Features:

• Standalone Training and Prediction Engine: The core engine can generate custom predictors using alternative training sets or validation schemas.
• Neural Network-Based Prediction: Uses neural networks to predict the pathological nature of single point amino acid mutations with an approximate success rate of 80% in humans.
• Performance Metrics: The default predictor reports Matthews Correlation Coefficient (MCC) values of 0.61 in 10-fold cross-validation and 0.42 on a blind test using SwissVar 2016 mutations.
• Mutational Hot Spot Scanning: Performs alanine scanning, massive mutation, and genetically accessible mutation procedures to identify critical regions within proteins.
• Hot Spot Profile Database: Contains a database of hot spot profiles for all non-redundant Protein Data Bank (PDB) structures.

Scientific Applications:

• Mendelian Variant Interpretation: Supports interpretation of pathogenic variants in Mendelian genetic disorders by predicting mutation pathogenicity.
• Disease Mechanism and Target Prioritization: Facilitates study of molecular disease mechanisms and prioritization of potential therapeutic targets through pathogenicity predictions.
• Protein Function and Residue Analysis: Enables analysis of protein function and identification of critical residues using mutational hot spot scanning.

Methodology:

Employs neural networks and allows training of custom predictors with alternative training sets and validation schemas; evaluates predictors with 10-fold cross-validation and blind testing reporting MCC; performs alanine scanning, massive mutation, and genetically accessible mutation scans and maintains hot spot profiles for non-redundant PDB structures.