Predicting antibody affinity changes upon mutations by combining multiple predictors.

Area of Science:

• Immunology
• Computational Biology
• Biochemistry

Background:

• Antibodies are key immune proteins with high affinity and specificity for antigens.
• Their use in biotherapeutics and bioengineering necessitates accurate prediction of affinity changes upon mutation.
• Current computational methods, including molecular mechanics and machine learning, lack sufficient accuracy for efficient antibody development.

Purpose of the Study:

• To develop a novel computational method for predicting antibody affinity changes upon mutation by integrating multiple predictors.
• To enhance the accuracy of antibody affinity prediction for improved antibody engineering.

Main Methods:

• A machine learning approach was employed to combine multiple antibody affinity prediction tools.
• The developed method was evaluated on the SiPMAB database.
• Performance was assessed using the Pearson's correlation coefficient (R) between predicted and experimental affinity changes ([Formula: see text]).

Main Results:

• The combined predictor method achieved a higher accuracy (R = 0.69) compared to individual molecular mechanics or machine learning methods (R = 0.59).
• The new method also surpassed previous approaches using the average of multiple predictors (R = 0.64).
• Feature importance analysis revealed that combining predictors with diverse strengths and protocols contributed to the improved accuracy.

Conclusions:

• Machine learning provides a powerful framework for integrating diverse computational approaches to predict antibody affinity changes.
• The developed ensemble method offers a more accurate solution for antibody engineering challenges.
• This study highlights the potential of combining computational strategies for advancing biotherapeutic development.