Digital Discovery of Synthesizable Metal−Organic Frameworks via Molecular Dynamics‑Informed, High‑Fidelity Deep Learning

Le résumé fourni par la source

Abstract Metal–organic frameworks (MOFs) are celebrated for their chemical and structural versatility, and in‑silico screening has significantly accelerated their discovery; yet most hypothetical MOFs (hMOFs) never reach the bench because their synthetic feasibility remains unknown. Herein, a deep‑learning proxy based on the Tabular Prior‐Data Fitted Network (TabPFN) is introduced that predicts free energy with high fidelity. The fine‐tuned surrogate attains a coefficient of determination ( R 2 ) of 0.96 and a mean absolute error (MAE) of 0.67 on the hold‑out test set, preserves strong temporal generalization ( R 2 = 0.89 and MAE = 1.35), as well as robust extrapolation on a chemically diverse blind set. Applying the surrogate to a well‐established hMOF library, we rapidly flag prime hMOF candidates that fall within the synthetic‑likelihood window observed for experimentally realized MOFs. This study delivers a practical bridge between digital reticular design and laboratory synthesis, transforming synthetic‑likelihood assessment from a computational bottleneck into a routine, high‑throughput step. Beyond MOFs, the approach provides a general framework for accelerating free‑energy‑driven materials discovery with minimal loss of accuracy, opening a new avenue to data‑driven exploration of complex chemical spaces.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.