MuSiC suite: PoPMuSiC, HoTMuSiC, and SNPMuSiC

MuSiC suite: PoPMuSiC, HoTMuSiC, and SNPMuSiC predict the effects of amino acid substitutions on protein folding stability and melting temperature to evaluate mutational impacts and classify genetic variants.

Key Features:

• Statistical potentials: The methods employ newly developed statistical potentials integrating four protein sequence and structure descriptors, including amino acid volume changes upon mutation.
• Energy function model: Predictions are computed as linear combinations of energy functions with coefficients determined by neural networks that account for solvent accessibility of mutated residues.
• Bias mitigation (PoPMuSiCsym): PoPMuSiCsym uses a balanced dataset (Ssym) and enforces physical symmetry under inverse mutations to reduce bias toward destabilizing mutations.
• Thermal stability prediction (HoTMuSiC): HoTMuSiC predicts mutation-induced changes in protein melting temperature from experimental or modeled structures using standard and temperature-dependent statistical potentials combined with artificial neural networks and was validated by 5-fold cross-validation.
• Disease variant classification (SNPMuSiC): SNPMuSiC classifies human variants as deleterious or neutral based on predicted stability changes using artificial neural networks and solvent accessibility-dependent statistical potentials, reporting a balanced accuracy of 71% and a positive predictive value of 89% in cross-validation.

Scientific Applications:

• Rational Protein Design: Predicts stability and thermal effects of substitutions to guide engineering of proteins and enzymes with altered properties.
• Protein Folding and Stability Insights: Provides quantitative estimates of folding free energy changes to inform studies of folding mechanisms and stability determinants.
• Genetic Variant Analysis: Classifies missense variants by their impact on protein stability to aid molecular interpretation of disease-associated mutations.

Methodology:

Statistical potentials (standard and temperature-dependent) integrating four sequence/structure descriptors including amino acid volume are combined linearly with coefficients trained by artificial neural networks that incorporate solvent accessibility; PoPMuSiCsym uses a balanced Ssym dataset and symmetry under inverse mutations; HoTMuSiC uses experimental or modeled structures and 5-fold cross-validation for validation; SNPMuSiC applies ANNs and solvent accessibility-dependent potentials for deleterious/neutral classification.