Topological Difference between Bound and Decoupled States Correlates with Protein–Ligand Binding Affinity

Abstract Accurate prediction of protein–ligand binding affinity is essential for drug discovery, yet understanding the underlying physical mechanisms remains a challenge. In this study, we employ persistent homology to compare experimental protein–ligand complexes with their hypothetical decoupled states, where intermolecular contacts are removed, and quantify the global topological difference between the two states by the Wasserstein distance. Analysis of nearly 20,000 complexes from the PDBbind v2020 dataset reveals that this single metric is negatively correlated with the experimental binding free energy, with the strongest Spearman correlation of approximately −0.37. Expanded into an element-specific feature vector, these topological difference features enhance the performance of a baseline model built on protein sequence embeddings and ligand descriptors. On the CASF-2016 benchmark, this increases the Pearson correlation coefficient from 0.844 to 0.866 and decreases the root mean square error from 1.643 to 1.556 kcal/mol, with permutation controls confirming that the gain reflects genuine structural information beyond added dimensionality. These results show that the bound-decoupled difference provides a topological perspective on protein–ligand binding affinity.