Based on geometric deep learning techniques, invariant point attention is used to predict the magnitude of protein complex binding affinity (Δ G) through three-dimensional characterization of protein complexes.