ABRefine: An Accurate Antibody Structure Refinement Method by Equivariant Graph Transformer with Rigid Body Constraint

Accurate antibody structural modeling is essential for therapeutic antibody development. However, even advanced artificial intelligence models such as AlphaFold2/3 struggle to precisely predict antibody structures due to the high flexibility of the complementarity-determining regions. Structure refinement methods can therefore play a critical role in improving the quality of predicted antibody conformations. In this paper, we present ABRefine, an end-to-end antibody structure refinement framework based on an equivariant graph transformer with rigid-body constraints. ABRefine directly predicts atomic coordinate shifts from initial structures while enforcing physically meaningful rigid-body consistency, enabling rapid and accurate refinement without conformational sampling. To our knowledge, ABRefine is also the first dedicated refinement method for nanobodies. Comprehensive evaluations on multiple antibody and nanobody benchmarks show that ABRefine outperforms existing refinement methods in both accuracy and efficiency, and notably enhances the structural quality of AlphaFold3 predictions.