A dataset of verified native protein heterodimer pairs and their cross-pair swaps,
derived from the NVIDIA AFDB collaboration (Han, Tsenkov, Venanzi et al., 2026).
Overview
Each row encodes two native heterodimer complexes from the same organism, both
verified by AlphaFold-Multimer with high structural confidence. From each row, a model
can construct 4 training instances: