Electron-affinity difference distributions as an organizing principle for superconductivity, enabling the discovery of PtPb₃Bi
Predicting superconducting transition temperature (Tc) remains challenging. This study proposes an interpretable, structure- and chemistry-aware Gaussian process model, GP-Tc, that encodes local bonding environments via graphlet histograms and constructs an effective kernel using Earth mover’s distance, enabling Tc prediction with uncertainty. Analysis shows that Tc across different superconducting families can be predicted using only the distribution of electron affinity differences between adjacent atoms, interatomic distances, and a small number of elemental features, revealing that electron affinity difference is a key chemical parameter linking local bonding to macroscopic superconductivity in a mechanism-agnostic manner. The model reproduces the experimental Tc range of the infinite-layer nickelate Nd0.8Sr0.2NiO2 and predicts superconductivity in stoichiometric PtPb3Bi, which is experimentally confirmed with Tc≈3 K. In addition, GP-Tc is openly accessible through a web interface and identifies high-priority candidate materials such as SrNiO2 and K(PRh)2.