SAAMBE-MEM: a sequence-based method for predicting binding free energy change upon mutation in membrane...

Area of Science:

• Biochemistry
• Computational Biology
• Structural Biology

Background:

• Mutations in protein-protein interactions can alter complex function and lead to disease.
• Assessing mutation impacts on membrane protein binding affinity is critical due to their abundance.
• Existing prediction methods often require structural data and are limited to soluble proteins.

Purpose of the Study:

• To develop a novel sequence-based method for predicting binding free energy changes (ΔΔG) in membrane protein-protein complexes due to mutations.
• To overcome limitations of existing methods that require structural information and are primarily trained on soluble proteins.

Main Methods:

• Developed SAAMBE-MEM, a sequence-based machine learning method using the MPAD database.
• Leveraged features like amino acid indices and Position-Specific Scoring Matrices (PSSM).
• Trained and validated the model using the XGBoost regression algorithm with curated datasets.

Main Results:

• SAAMBE-MEM achieved a Pearson correlation coefficient of 0.64 with optimal PSSM-related features.
• The method outperformed existing approaches trained on the SKEMPI database.
• Evolution-based features demonstrated superior performance compared to physicochemical features.

Conclusions:

• SAAMBE-MEM provides an effective sequence-based approach for predicting mutation effects on membrane protein binding affinity.
• The method's reliance on evolution-based features highlights their importance in understanding these interactions.
• SAAMBE-MEM is accessible via a web server and standalone code.