Separating selection from mutation in antibody language models.

Area of Science:

• Immunology
• Computational Biology
• Protein Engineering

Background:

• Antibodies evolve through V(D)J recombination, mutation, and selection.
• Current antibody language models focus on amino acid sequences, implicitly learning mutation processes.
• This implicit learning degrades performance in predicting functional effects of mutations.

Purpose of the Study:

• To develop a novel framework for antibody language modeling that explicitly separates mutation and selection.
• To improve the prediction of functional effects of amino acid mutations in antibodies.
• To create a more efficient and interpretable antibody modeling approach.

Main Methods:

• Devised a Deep Amino acid Selection Model (DASM) framework.
• Explicitly factored out nucleotide-level mutation processes.
• Fitted selection effects as a separate term from mutation processes.

Main Results:

• DASM substantially improved performance on standard antibody functional benchmarks.
• The model exclusively quantifies functional effects of mutations.
• Achieved improved prediction of mutation effects on antibody function.

Conclusions:

• Separating mutation and selection processes in antibody language models enhances functional prediction.
• DASM offers a more accurate and efficient approach to antibody engineering.
• The model's interpretability aids in understanding antibody evolution and function.