Evaluation of AlphaFold Antibody-Antigen Modeling with Implications for Improving Predictive Accuracy

Area of Science:

• Structural Biology
• Immunology
• Computational Biology

Background:

• High-resolution antibody-antigen structures are crucial for understanding immune recognition and developing therapeutics.
• Experimental structure determination is challenging, necessitating accurate computational modeling tools for antibody-antigen complexes.
• Previous versions of AlphaFold demonstrated limited success in modeling antibody-antigen interactions compared to general protein-protein complexes.

Approach:

• Analyzed AlphaFold's performance on 429 nonredundant antibody-antigen complex structures.
• Identified confidence metrics for predicting model quality.
• Investigated complex features associated with improved modeling success.

Key Points:

• AlphaFold's current version improves near-native antibody-antigen model success to over 30%, a significant increase from approximately 20% in previous versions.
• The study highlights the importance of bound-like component modeling for accurate complex assembly.
• Useful confidence metrics and specific complex features associated with improved modeling were identified.

Conclusions:

• AlphaFold can now generate accurate antibody-antigen models in many cases.
• Further training of AlphaFold holds potential for enhanced performance in antibody-antigen complex modeling.
• The findings support the utility of AlphaFold as a valuable tool in structural immunology and therapeutic design.