Predicting adsorption ability of adsorbents at arbitrary sites for pollutants using deep transfer learning

Accurately evaluating the adsorption ability of adsorbents for heavy metal ions (HMIs) and organic pollutants in water is critical for designing efficient adsorbents, but predicting adsorption at arbitrary sites is challenging due to the lack of measuring technologies for active sites and activities. This work presents an AI approach that uses a deep neural network and transfer learning (TL) to predict adsorption abilities at arbitrary sites, demonstrated for three HMIs (Pb(II), Hg(II), Cd(II)) on two-dimensional graphitic-C3N4 (g-C3N4). A DNN trained on 7000 DFT-calculated adsorption energies for Pb(II)/g-C3N4 serves as the source model. With TL, adsorption capabilities for Hg(II) and Cd(II) are accurately predicted using only 700 data points each, achieving RMSEs less than 0.1 eV and the order Cd(II) > Hg(II) > Pb(II). The AI method attains DFT-level accuracy while being millions of times faster in prediction and requiring one-tenth the data compared to training from scratch. Experimental measurements of g-C3N4 adsorption capacities for the three HMIs corroborate the AI predictions, indicating broad applicability of the approach to other adsorbents and pollutants.

Zhilong Wang, Haikuo Zhang, Jiahao Ren, Xirong Lin, Tianli Han, Jinyun Liu, Jinjin Li