Massive Sampling Strategy for Antibody-Antigen Targets in CAPRI Round 55 With MassiveFold

Area of Science:

• Computational biology
• Structural biology
• Biophysics

Background:

• AlphaFold2 has advanced protein structure prediction.
• Previous methods like AFsample showed promise in CASP15-CAPRI.
• Antibody-antigen interactions remain a significant challenge for prediction tools.

Purpose of the Study:

• To evaluate the effectiveness of massive sampling with AlphaFold2 for antibody-antigen complex prediction.
• To assess the utility of AlphaFold2 confidence scores in identifying high-quality models.
• To compare different sampling strategies for optimizing predictions.

Main Methods:

• Utilized AlphaFold2-based MassiveFold, generating over 6000 predictions per target across 6 pools.
• Applied distinct parameter sets for each prediction pool.
• Focused on antibody-antigen targets from CAPRI Round 55.

Main Results:

• Massive sampling consistently yielded acceptable to high-quality predictions.
• AlphaFold2's internal confidence score was not a reliable indicator of model accuracy.
• Increased sampling without dropout outperformed other strategies for most targets.

Conclusions:

• Massive sampling is a viable strategy for improving protein-protein complex predictions, including difficult antibody-antigen cases.
• Relying solely on AlphaFold2 confidence scores can be misleading.
• Optimizing sampling parameters, particularly increasing sampling without dropout, is crucial for achieving top-tier predictions.