Integrating Dynamic Network Analysis with AI for Enhanced Epitope Prediction in PD-L1:Affibody Interactions

Area of Science:

• Computational structural biology
• Biophysics
• Protein-protein interactions

Background:

• Determining protein-protein interaction (PPI) binding epitopes is crucial for drug discovery and biomedicine.
• Experimental epitope determination methods are often low-throughput and expensive.
• Current artificial intelligence (AI) methods, including AlphaFold2, struggle with predicting epitopes for novel binding domains.

Purpose of the Study:

• To develop an integrated computational method for accurate prediction of protein binding epitopes.
• To unravel the structure and binding epitope of the PD-L1:Affibody complex.
• To assess the performance of AI models and conventional methods in predicting PPI interfaces.

Main Methods:

• Integrated generalized-correlation-based dynamic network analysis (GCCA-DyNA) on multiple molecular dynamics (MD) trajectories.
• Initiation of MD simulations from AlphaFold2Multimer predicted structures.
• Validation using experimental epitope mapping techniques: cross-linking mass spectrometry (XL-MS) and deep mutational scanning (DMS).

Main Results:

• The integrated GCCA-DyNA method successfully distinguished between parallel and perpendicular binding models for the PD-L1:Affibody complex, identifying the perpendicular mode as significantly more stable.
• Experimental validation confirmed the computational predictions of the perpendicular binding pose.
• AlphaFold3 failed to predict the correct binding pose, highlighting limitations in current AI structure prediction for novel interactions.

Conclusions:

• Dynamic network analysis integrated with AI-predicted structures offers a robust approach for accurate epitope identification.
• The findings challenge the uncritical acceptance of AI-generated protein structures and emphasize the need for complementary computational strategies.
• This integrated method enhances the reliability of predicting protein-protein interaction interfaces, with implications for therapeutic development.