Prot-ΔΔG: Prediction of protein-protein binding affinity changes upon mutations with pre-training strategies

Area of Science:

• Biochemistry
• Computational Biology
• Genetics

Background:

• Protein-protein interactions are vital for biological functions.
• Amino acid mutations can disrupt these interactions, causing disease.
• Quantifying mutation impact via binding affinity change (ΔΔG) is essential for research.

Purpose of the Study:

• To develop an accurate computational method for predicting ΔΔG using only protein sequences.
• To overcome limitations of structure-dependent methods in predicting mutation impacts.
• To provide a broadly applicable tool for understanding disease mechanisms and protein engineering.

Main Methods:

• Developed Prot-ΔΔG, a deep learning framework utilizing pre-trained protein language models.
• Integrated a BiGRU-DBRNN encoder to process sequence-embedded patterns and evolutionary information.
• Leveraged only wild-type and mutant amino acid sequences, eliminating the need for structural data.

Main Results:

• Prot-ΔΔG demonstrated competitive performance in predicting ΔΔG for single, mixed, and multi-point mutations.
• Significant performance improvement was observed in protein-level blind testing.
• The sequence-based approach proved effective in capturing complex biological patterns.

Conclusions:

• Prot-ΔΔG offers a robust, sequence-only method for predicting mutation effects on protein-protein binding affinity.
• This approach enhances applicability, particularly when protein structures are unavailable or unreliable.
• The findings advance computational prediction in molecular biology, drug discovery, and protein engineering.