Ibex: Pan-immunoglobulin structure prediction

We introduce Ibex, an immunoglobulin protein structure prediction model that achieves state-of-the-art accuracy in modeling the binding domains of antibodies, nanobodies and T-cell receptors. Ibex can model both bound and unbound conformations of the protein, having been trained on labeled apo and holo pairs. Using a private dataset of hundreds of antibody structures, we evaluate the out-of-distribution performance of common structure prediction tools, showing improved robustness of Ibex compared to existing specialized structure prediction models.