Enhancing antibody-antigen interaction prediction with atomic flexibility.

Area of Science:

• Immunology and Computational Biology
• Protein Structure and Dynamics

Background:

• Antibodies are key immune components essential for vaccines and therapeutics.
• Modeling antibody-antigen interactions is complex due to antibody flexibility and dynamic binding.
• Deep learning has advanced protein structure prediction but struggles with antibody-antigen dynamics.

Purpose of the Study:

• To investigate the use of predicted Local Distance Difference Test (pLDDT) scores to model antibody flexibility.
• To enhance computational modeling of antibody-antigen (Ab-Ag) interactions.
• To improve the engineering of antibodies for enhanced affinity and breadth against variable pathogens.

Main Methods:

• Utilized a fingerprint-based approach incorporating pLDDT scores as indicators of residue flexibility.
• Examined the impact of flexibility on antibody-specific tasks and Ab-Ag interaction modeling.
• Evaluated model performance using Area Under the Receiver Operating Characteristic Curve (AUC-ROC).

Main Results:

• Incorporating flexibility through pLDDT scores enhanced Ab-Ag interaction model predictive accuracy by 4%, achieving an AUC-ROC of 92%.
• Demonstrated state-of-the-art performance in paratope prediction.
• pLDDT scores were shown to be a valuable proxy for conformational flexibility in antibody modeling.

Conclusions:

• Accounting for conformational flexibility is critical for accurate antibody-antigen interaction modeling.
• pLDDT scores offer a practical method to represent and optimize antibody flexibility.
• This approach holds significant promise for engineering antibodies against highly variable pathogens like HIV and SARS-CoV-2.