DDMut-PPI

DDMut-PPI predicts changes in protein-protein interaction binding free energy (ΔΔG) caused by single and multiple point mutations using deep learning to quantify mutational effects on PPIs.

Key Features:

• Deep Learning Architecture: Siamese network architecture enhanced with graph convolutional networks (GCNs) focused on protein interaction interfaces.
• Graph-Based Signatures: Represents PPI interfaces with graph-based signatures using residue-specific embeddings from the ProtT5 protein language model as node features and molecular interactions as edge features.
• Context Integration: Integrates evolutionary context with spatial interface information to maintain consistent accuracy across mutations that increase or decrease binding affinity.
• Performance Metrics: Reported Pearson correlation up to 0.75 and root mean squared error (RMSE) of 1.33 kcal/mol for predicted ΔΔG, outperforming most existing methods in evaluations.

Scientific Applications:

• Cellular function analysis: Predicts how point mutations alter PPI binding free energy to study effects on cellular processes mediated by protein interactions.
• Therapeutic development: Assesses mutation impacts relevant for designing or optimizing protein therapeutics and interventions targeting PPIs.
• Disease mechanism investigation: Evaluates mutation-driven perturbations of PPIs to elucidate molecular mechanisms underlying disease.

Methodology:

Siamese network architecture combined with graph convolutional networks operating on the protein interaction interface, using graph-based signatures with ProtT5 residue embeddings as node features and molecular interactions as edge features, and integrating evolutionary and spatial data to predict ΔΔG.