PackPPI: An integrated framework for protein-protein complex side-chain packing and ΔΔG prediction based on diffusion...

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Engineering

Background:

• Deep learning advances protein side-chain packing and mutation effect prediction (ΔΔG).
• These related tasks are often handled separately, limiting integrated solutions.
• Existing methods lack effective post-processing, hindering conformational refinement.

Purpose of the Study:

• Introduce PackPPI, an integrated framework for protein complex analysis.
• Improve side-chain prediction accuracy and ΔΔG prediction using learned representations.
• Enhance conformational plausibility through effective post-processing.

Main Methods:

• Utilizes a diffusion model for generating protein conformations.
• Employs a proximal optimization algorithm for refining side-chain packing.
• Integrates learned representations for predicting mutation-induced binding affinity changes (ΔΔG).

Main Results:

• Achieved lowest atom RMSD (0.9822) on the CASP15 dataset for side-chain prediction.
• Proximal optimization effectively reduced spatial clashes while maintaining a low-energy landscape.
• Demonstrated state-of-the-art performance in predicting binding affinity changes on SKEMPI v2.0.

Conclusions:

• PackPPI offers a robust and versatile computational tool for protein design and engineering.
• The integrated approach enhances accuracy in both conformational prediction and ΔΔG estimation.
• The framework shows significant potential for advancing molecular modeling and drug discovery.