Machine-Learning-Assisted Material Discovery of Pyridine-Based Polymers for Efficient Removal of ReO4.

Abstract

Efficient capture of ⁹⁹TcO₄^- is the focus in nuclear waste management. For laboratory operation, ReO₄^- is used as a nonradioactive alternative to ⁹⁹TcO₄^- to develop high-performance adsorbents for the treatment. However, the traditional design of new adsorbents is primarily driven by the chemical intuition of scientists and experimental methods, which are inefficient. Herein, a machine learning (ML)-assisted material genome approach (MGA) is proposed to precisely design high-efficiency adsorbents. ML models were developed to accurately predict adsorption capacity from adsorbent structures and solvent environment, thus predicting and screening the 2450 virtual pyridine polymers obtained by MGA, and it was found that halogen functionalization can enhance its adsorption efficiency. Two halogenated functional pyridine polymers (F-C-CTF and Cl-C-CTF) predicted by this approach were synthesized that exhibited excellent acid/alkali resistance and selectivity for ReO₄^-. The adsorption capacity reached 940.13 (F-C-CTF) and 732.74 mg g^-1 (Cl-C-CTF), which were better than those of most reported adsorbents. The adsorption mechanism is comprehensively elucidated by experiment and density functional theory calculation, showing that halogen functionalization can form halogen-bonding interactions with ⁹⁹TcO₄^-, which further justified the theoretical plausibility of the screening results. Our findings demonstrate that ML-assisted MGA represents a paradigm shift for next-generation adsorbent design.