1. Molecular fingerprint and machine learning enhance high-performance MOFs for mustard gas removal
- Author
-
Jing Ni, Jinfeng Li, Shuhua Li, He Zheng, Zhongyuan Ming, Li Li, Heguo Li, Shouxin Zhang, Yue Zhao, Hong Liang, and Zhiwei Qiao
- Subjects
Natural sciences ,Chemistry ,Materials science ,Science - Abstract
Summary: Chemical warfare agents (CWAs), epitomized by the notoriously used mustard gas (HD), represent a class of exceptionally toxic chemicals whose airborne removal is paramount for battlefield safety. This study integrates high-throughput computational screening (HTCS) with advanced machine learning (ML) techniques to investigate the efficacy of metal-organic frameworks (MOFs) in adsorbing and capturing trace amounts of HD present in the air. Our approach commenced with a comprehensive univariate analysis, scrutinizing the impact of six distinct descriptors on the adsorption efficiency of MOFs. This analysis elucidated a pronounced correlation between MOF density and the Henry coefficient in the effective capture of HD. Then, four ML algorithms were employed to train and predict the performance of MOFs. The Random Forest (RF) algorithm demonstrates strong model learning and good generalization, achieving the best prediction result of 98.3%. In a novel exploratory stride, we incorporated a 166-bit MACCS molecular fingerprinting (MF) to identify critical functional groups within adsorbents. From the top 100 MOFs analyzed, 22 optimal functional groups were identified. Leveraging these insights, we designed three innovative substructures, grounded in these key functional groups, to enhance HD adsorption efficiency. In this work, the combination of MF and ML could provide a new direction for efficient screening of MOFs for the capture of HD in the air. The outcomes of this study offer substantial potential to revolutionize the domain of CWA capture. This represents a significant stride toward developing practical solutions that enhance both environmental protection and battlefield security.
- Published
- 2024
- Full Text
- View/download PDF