Enhancing Superconductor\nCritical Temperature Prediction:\nA Novel Machine Learning Approach Integrating Dopant Recognition

Doping plays a crucial role in determining the critical\ntemperature\n(<i>T</i><sub>c</sub>) of superconductors, yet accurately\npredicting its effects remains a significant challenge. Here, we introduce\na novel doping descriptor that captures the complex influence of dopants\non superconductivity. By integrating the doping descriptor with elemental\nand physical features within a Mixture of Experts (MoE) model, we\nachieve a remarkable <i>R</i><sup>2</sup> of 0.962 for <i>T</i><sub>c</sub> prediction, surpassing all published prediction\nmodels. Our approach successfully identifies optimal doping levels\nin the Bi<sub>2–<i>x</i></sub>Pb<sub><i>x</i></sub>Sr<sub>2</sub>Ca<sub>2–<i>y</i></sub>Cu<sub><i>y</i></sub>O<sub><i>z</i></sub> system, with\npredictions closely aligning with experimental results. Leveraging\nthis model, we screen compounds from the Inorganic Crystal Structure\nDatabase and employ a generative approach to explore new doped superconductors.\nThis process reveals 40 promising candidates for high <i>T</i><sub>c</sub> superconductivity among existing and hypothetical doped\nmaterials. By explicitly accounting for doping effects, our method\noffers a powerful tool for guiding the experimental discovery of new\nsuperconductors, potentially accelerating progress in high-temperature\nsuperconductivity research and opening new avenues for material design.