FFMSRes-MutP

FFMSRes-MutP predicts disease-associated non-synonymous single nucleotide polymorphisms (nsSNPs) from biological data to prioritize pathogenic variants.

Key Features:

• Multi-Scale ResNet Models: Builds upon MSRes-MutP by integrating multi-scale kernel sizes within ResNet blocks to capture complex patterns in biological data.
• Deep Feature Fusion Strategy: Employs a deep feature fusion strategy to combine two-dimensional features and physicochemical properties extracted using both 2D-ResNet and 1D-ResNet blocks.
• Four Feature Types Extracted: Extracts four types of features from biological data for comprehensive representation (feature types as reported in the source).
• Enhanced Performance Metrics: Achieves MCCs of 0.593 (PredictSNP) and 0.618 (MMP), outperforming the previous best method by 0.101 and 0.210 respectively, and records MCCs of 0.9605 (HumDiv) and 0.9507 (HumVar) with AUCs of 0.9796 and 0.9748.
• Robust Benchmarking: Validated on five datasets and ranked as the second-best predictor in an independent blind test with MCC 0.5215 and AUC 0.7633.

Scientific Applications:

• Genetic research: Supports analysis of molecular mechanisms underlying nsSNP-associated diseases by prioritizing pathogenic variants.
• Drug discovery: Identifies candidate variants and associated molecular features that can inform potential therapeutic targets.
• Personalized medicine: Aids prioritization of clinically relevant nsSNPs to support variant interpretation and individualized therapeutic strategies.

Methodology:

Extracts four types of features from biological data and processes them via a deep feature fusion strategy leveraging 2D-ResNet and 1D-ResNet blocks.