Electron affinity difference distributions guide the discovery of the superconductor PtPb$_3$Bi

Predicting the superconducting transition temperature ($T_c$) from crystal structure and composition remains a central challenge in condensed-matter physics, reflecting the absence of a broadly predictive framework connecting microscopic bonding to macroscopic quantum behavior. Here, we introduce a structure- and chemistry-aware approach implemented in an interpretable Gaussian process model, which we call GP-$T_c$ (Gaussian Process $T_c$), that enables uncertainty-quantified prediction of superconductivity from experimentally accessible inputs. By encoding local bonding environments and geometry as graphlet histograms and learning within a probabilistic framework, we find that the predictive space collapses to a compact set of descriptors: the distribution of electron-affinity differences between neighboring atoms, together with simple elemental features and interatomic distances, provides an informative basis for predicting $T_c$ across disparate superconducting families. This result identifies an overlooked chemical control parameter while emphasizing the essential role of local structure beyond composition-only approaches. We demonstrate the framework through two complementary tests: validation against a recently established superconducting family and discovery of a previously unknown material. GP-$T_c$ reproduces the experimentally reported $T_c$ range of the infinite-layer nickelate Nd0.8Sr0.2NiO2. We further predict superconductivity in stoichiometric PtPb$_3$Bi and experimentally confirm it through synthesis and bulk measurements, establishing PtPb$_3$Bi as a new superconductor with $T_c$~3 K. GP-$T_c$ identifies additional high-priority superconducting candidates -- including SrNiO2, K(PRh)2, and Ho2C3 -- that provide concrete targets for ongoing and future experimental exploration.