AlphScore

AlphScore predicts the pathogenicity of missense variants by integrating structural features derived from AlphaFold2 protein models with machine learning classification.

Key Features:

• AlphaFold2 Structural Feature Integration: Extracts structural features from AlphaFold2-predicted protein models at each amino acid position.
• Solvent Accessibility Analysis: Evaluates residue solvent exposure to assess potential functional effects of missense variants.
• Amino Acid Network Features: Analyzes residue interaction networks within protein structures to detect disruptions caused by mutations.
• Physicochemical Environment Descriptors: Characterizes local chemical environments surrounding amino acid residues to model mutation effects.
• pLDDT Confidence Incorporation: Uses AlphaFold2 predicted local distance difference test (pLDDT) scores as structural confidence indicators.
• Random Forest Classification Model: Applies a random forest algorithm trained on missense variants from gnomAD v3.1 to distinguish proxy-benign and proxy-pathogenic variants.

Scientific Applications:

• Missense Variant Pathogenicity Prediction: Supports computational classification of missense variants using structural protein features.
• Genomic Variant Interpretation: Improves interpretation of human genetic variants through integration with predictive scores such as CADD and REVEL.
• Functional Genomics Analysis: Facilitates evaluation of structural impacts of amino acid substitutions on protein function.

Methodology:

AlphScore extracts solvent accessibility, amino acid network features, physicochemical environment descriptors, and pLDDT confidence scores from AlphaFold2 protein structures and trains a random forest classifier on gnomAD v3.1 missense variants to distinguish proxy-benign and proxy-pathogenic variants.