Towards a Greener AlphaFold2 Protocol for Antibody-Antigen Modeling: Insights From CAPRI Round 55

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• The Critical Assessment of PRediction of Interactions (CAPRI) challenge benchmarks protein structure prediction methods.
• Accurate prediction of protein-protein interactions, especially antibody-antigen complexes, remains a significant challenge in structural biology.

Purpose of the Study:

• To evaluate an enhanced AlphaFold2 (AF2) sampling protocol combined with data-driven docking for predicting protein complexes in the CAPRI 55th round.
• To assess the performance of a reduced MinnieFold sampling strategy and the FoldX scoring function in complex prediction.

Main Methods:

• Utilized an enhanced AlphaFold2 (AF2) protocol based on Wallner's massive sampling approach, generating thousands of models per target.
• Employed a 50-fold reduced MinnieFold sampling and custom ranking for specific targets (T231, T232).
• Applied data-driven docking for antibody-MHC I complexes (T233, T234) and used FoldX for model scoring.

Main Results:

• Achieved top-ranking medium predictions for the antibody-peptide complex (T231) and PP2A:TIPRL complex (T232) using MinnieFold sampling.
• Data-driven docking failed to produce acceptable models for antibody-MHC I complexes (T233, T234).
• Post-analysis suggested MinnieFold could have yielded a medium-quality model for T233, and FoldX successfully selected acceptable/medium-quality models for T231 and T232.

Conclusions:

• Enhanced AF2 sampling, particularly the MinnieFold approach, demonstrates feasibility for accurate and efficient prediction of antibody complexes.
• The findings highlight the potential of green and accurate enhanced AF2 sampling strategies for antibody complex modeling.
• Custom ranking and scoring functions like FoldX are crucial for identifying high-quality predictions from large model sets.