Use of Antibody Structural Information in Disease Prediction Models Reveals Antigen Specific B Cell Receptor Sequences in Bulk Repertoire Data

Convergent antibodies are highly similar antibodies elicited in multiple individuals in response to the same antigen. Convergent antibodies provide insight into shared immunological responses and show great promise as diagnostic biomarkers. They have typically been identified using methods that consider the amino acid sequence of the third complementarity-determining region (CDR3) of immunoglobulin heavy chain (IgH). In this study, we extend this definition of convergent antibodies to use structural information about the three IgH CDR regions (CDR1-3) and antibody isotype. We measure the performance of both definitions of convergence by their ability to predict disease status from bulk IgH sequencing data for two different disease cohorts: HIV infection and food allergen sensitization. We show that using predicted structural information outperforms prior approaches for the prediction of food sensitization status and performs on par for HIV infection status. Additionally, the use of structural information and antibody isotype enables better model interpretation and more immunological insights. The structurally convergent antibody groups driving HIV prediction accuracy are from known HIV-binding antibodies expressing the appropriate isotype. Thus, the incorporation of structural information can enable the identification of antigen specific antibody groups from bulk IgH sequencing data.