ProS-GNN: Predicting effects of mutations on protein stability using graph neural networks

Area of Science:

• Computational biology
• Protein engineering
• Bioinformatics

Background:

• Predicting protein stability changes upon mutation is crucial for understanding drug failure mechanisms and developing immunotherapies.
• Existing machine learning models often exhibit biases or poor generalization for predicting mutation effects on protein stability.

Purpose of the Study:

• To develop an advanced computational approach for accurately predicting protein stability changes caused by mutations.
• To overcome the limitations of existing methods, specifically anti-symmetric bias and poor generalization.

Main Methods:

• A gated graph neural network (GNN) model was developed using message passing to encode molecular structure-property relationships.
• Input feature vectors were created by eliminating non-mutant structures and incorporating raw atomic coordinates for spatial insights.
• The model was evaluated on diverse datasets including S^sym, Myoglobin, Broom, and p53.

Main Results:

• The proposed gated GNN approach demonstrated improved generalization performance across multiple datasets.
• The method achieved better linearity and symmetry in predictions compared to existing computational approaches.
• The prediction process was completed in less time than conventional methods.

Conclusions:

• The novel gated graph neural network provides a more accurate and efficient computational tool for predicting protein stability changes upon mutation.
• This approach offers a valuable advancement for drug development and immunotherapy research by reliably predicting mutation impacts.
• The developed model shows superior performance and generalization capabilities, addressing key limitations in current predictive techniques.