SFM-Net: Selective Fusion of Multiway Protein Feature Network for Predicting Binding Affinity Changes upon Mutations

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Genomics

Background:

• Predicting mutation effects on protein-protein interactions (PPIs) is vital for understanding protein function and disease mechanisms.
• Current methods often struggle to fully utilize complex structural information and integrate diverse data sources.
• Accurate prediction of binding affinity changes (ΔΔG) due to mutations is a significant challenge.

Purpose of the Study:

• To develop a novel deep learning model, SFM-Net, for accurate prediction of mutation-induced binding affinity changes.
• To effectively leverage sequence, structural, and evolutionary information for enhanced prediction accuracy.
• To address the challenge of integrating multisource features in mutation effect prediction.

Main Methods:

• Developed SFM-Net, a deep learning model incorporating Graph Neural Network (GNN)-based multiway feature extractors.
• Implemented a context-aware selective fusion module to jointly utilize sequence, structural, and evolutionary data.
• Employed benchmarking experiments and ablation studies to validate the model's performance.

Main Results:

• SFM-Net demonstrates effective and selective integration of features from multiple sources.
• The model shows improved accuracy in predicting binding affinity changes (ΔΔG) caused by mutations.
• Ablation studies confirm the robustness and effectiveness of the proposed method.

Conclusions:

• SFM-Net offers a powerful approach for predicting the impact of mutations on PPIs.
• The model's ability to selectively fuse diverse data sources enhances prediction accuracy.
• This work contributes to a better understanding of mutation effects in protein structure and function.