A Data-Driven Approach to Antigen-Antibody Complex Structure Modeling Using Labeled VHH Antibodies

Abstract

Tumor necrosis factor alpha (TNF$\alpha$) has been extensively studied using X-ray crystallography, cryo-electron microscopy, and AI-based modeling. Antibodies have also been used as structural probes to investigate TNF$\alpha$. To enrich antigen-antibody structural data, we immunized alpacas with human TNF$\alpha$ and developed a VHH (single domain antibody) library. VHH antibodies consist of a single chain, which facilitates large-scale data collection. However, accurately modeling antigen-antibody complexes in three-dimensional (3D) remains a challenge. We selected TNF$\alpha$-binding VHH clones and predicted their 3D structures using ColabFold. By adding our VHH library, we generated multiple sequence alignments (MSAs) with much higher sequence coverage than those using only public databases. The predicted local distance difference test (pLDDT) of structural models for multiple outputs tended to be better, suggesting the usefulness of our antibody library. In this study, we released a labeled antibody dataset for TNF$\alpha$ and attempted structural modeling, which we hope will facilitate advances in antibody engineering and therapeutics.