AI-accelerated discovery of altermagnetic metal-organic frameworks via interpretable machine learning

Abstract

Altermagnetism has emerged as a third class of collinear magnetism, yet its exploration remains largely confined to inorganic materials. Here, we establish an AI-accelerated pipeline for discovering altermagnetic metal-organic frameworks (MOFs). From symmetry-guided DFT screening of 350 candidates, we identified six altermagnetic MOFs. A two-stage machine learning strategy was then employed: First, an XGBoost classifier pre-screened 65,578 candidates, narrowing the pool to 145 high-potential linkers, from which DFT confirmed six new altermagnets. Building on this enriched dataset, we trained a KNN classifier (F1 = 0.8793) and a CatBoost regressor (R² = 0.7732) for predicting altermagnetism and spin-splitting magnitudes, respectively. SHAP analysis translated predictions into design principles: conjugated backbones favor altermagnetism, while moderate push-pull groups enhance spin splitting. Guided by these principles, we designed three additional MOF candidates, all confirmed by DFT. In total, this workflow identified 15 altermagnetic MOFs, establishing a generalizable framework for accelerating quantum materials discovery.