MOVA

MOVA evaluates the pathogenic potential of missense variants by leveraging AlphaFold2 structural predictions to incorporate positional information into variant pathogenicity assessments, with a focus on amyotrophic lateral sclerosis (ALS).

Key Features:

• Integration with AlphaFold2: Uses AlphaFold2 predicted protein 3D structures to place missense variants within a structural context.
• Positional Information Utilization: Analyzes variant x, y, and z coordinates to identify structural regions and hotspots where pathogenic variants concentrate in ALS-related genes.
• Feature Analysis and Machine Learning: Trains a random forest on features including x, y, z coordinates, pLDDT scores, and BLOSUM62 substitution matrices, evaluated with stratified fivefold cross-validation.
• Comparative Performance Evaluation: Benchmarks predictive performance against other in silico methods across 12 ALS-related genes, including TARDBP, FUS, and SOD1.
• Enhanced Predictive Accuracy: Reports AUC ≥ 0.70 for multiple genes and shows improved accuracy when combined with REVEL or CADD.
• Impactful Features: Identifies spatial coordinates (x, y, z) as among the most predictive features of pathogenicity.

Scientific Applications:

• ALS variant interpretation: Prioritizes rare missense variants in ALS genes for pathogenicity assessment.
• Neurodegenerative disease research: Investigates how structural changes from missense mutations may contribute to disease mechanisms in neurodegeneration.
• Variant prioritization for follow-up: Supports selection of candidate variants for therapeutic targeting or further genetic study.

Methodology:

Uses AlphaFold2-predicted 3D structures and features (variant x, y, z coordinates, pLDDT scores, BLOSUM62) to train a random forest classifier evaluated by stratified fivefold cross-validation and benchmarks performance across 12 ALS-related genes (e.g., TARDBP, FUS, SOD1), reporting AUC values and combined analyses with REVEL or CADD.