AbCVista: a deep learning framework for predicting antibody conformational ensembles

Predicting antibody structures is crucial in biomedical research, particularly for developing antibody-based therapeutics and elucidating immune responses. However, the prevailing paradigm in protein structure prediction—producing a single, high-accuracy static model—falls short for antibodies. It struggles not only to accurately model the hypervariable complementarity-determining region (CDR) loops but also to adequately represent the functionally critical conformational ensembles of these regions. Here, we present AbCVista, an antibody-specific structure-prediction framework that outperforms state-of-the-art baselines in accuracy and physical plausibility while generating meaningful conformational ensembles that capture the dynamics of key binding regions. This capability overcomes a major limitation in characterizing antibody flexibility and shifts the field from static snapshots to a quantifiable, dynamic view of the structural landscape. By providing reliable ensembles rather than single models, AbCVista enables more robust developability risk assessment and has the potential to advance affinity maturation and antigen–antibody interaction analysis, accelerating the discovery and optimization of antibody therapeutics. The online service of AbCVista is freely available at http://cao.labshare.cn/abcvista/.