MuPIPR

MuPIPR predicts the effects of point mutations on protein-protein interactions using an end-to-end deep learning framework.

Key Features:

• End-to-end deep learning framework: Maps sequence-level mutation information directly to quantitative changes in PPI properties through joint model training.
• Contextualized amino-acid representations: Propagates the effect of a point mutation throughout surrounding amino acid representations to amplify subtle changes within long protein sequences.
• Siamese residual recurrent convolutional neural encoder: Encodes wildtype and mutant protein pairs with a Siamese architecture combining residual, recurrent, and convolutional components.
• Multiple-layer perceptron regressors: Applies MLP regressors to encoded representations to predict quantifiable changes in PPI properties.
• Targets PPI metrics: Produces predictions for changes in binding affinity and buried surface area as measures of mutation impact.
• Captures biophysical effects: Designed to reflect alterations in conformation and thermodynamics of protein-protein interactions induced by mutations.
• Experimental performance: Demonstrated improved performance over various state-of-the-art systems in predicting changes in binding affinity and buried surface area.

Scientific Applications:

• Mutation effect prediction: Estimating how point mutations alter protein-protein interaction strength and interface properties.
• Binding affinity analysis: Predicting mutation-induced changes in binding affinity between protein partners.
• Interface geometry assessment: Predicting changes in buried surface area to assess alteration of interaction interfaces by mutations.
• Biophysical impact analysis: Investigating mutation-driven conformational and thermodynamic changes in PPIs.

Methodology:

MuPIPR uses contextualized amino-acid representations that propagate mutation effects, encodes wildtype and mutant protein pairs with a Siamese residual recurrent convolutional neural encoder, and applies multiple-layer perceptron regressors to the encoded representations to predict quantitative changes in PPI properties.