Benchmarking antibody clustering methods using sequence, structural, and machine learning similarity measures for...

Area of Science:

• Biochemistry
• Immunology
• Bioinformatics

Background:

• Antibodies are crucial biotherapeutics derived from immune system proteins.
• Discovering therapeutic antibodies involves analyzing vast sequence data from methods like phage display.
• Current methods often rely solely on sequence similarity for grouping antibody candidates, potentially limiting diversity.

Purpose of the Study:

• To benchmark diverse antibody grouping methods beyond sequence similarity.
• To evaluate grouping performance on binder detection and epitope mapping tasks.
• To assess the diversity gains from combining orthogonal grouping strategies.

Main Methods:

• Benchmarking antibody grouping using clonotype, sequence, paratope prediction, structure prediction, and embedding.
• Evaluating methods on binder detection and epitope mapping.
• Developing an online tool (CLAP) for visualizing and contrasting antibody groupings.

Main Results:

• No single method significantly outperformed others in binder detection.
• Clonotype, paratope, and embedding-based groupings were top performers for epitope mapping.
• All tested methods provided orthogonal groupings, enhancing candidate pool diversity when used in combination.

Conclusions:

• Combining multiple antibody grouping methods increases candidate pool diversity.
• The CLAP tool facilitates exploration and visualization of antibody diversity across different grouping strategies.
• This approach aids in selecting a more diverse set of antibody candidates for therapeutic development.