Scholarlink

<p>The structure prediction of metal-organic frameworks (MOFs) prior to synthesis is crucial for realizing rational discovery of materials with customized properties, while the extensive trial-and-error experimentation could be substantially reduced. However, the diversity of metal-containing secondary building units (SBUs) poses a major challenge to predicting the underlying net architectures from scratch. In this work, we developed an efficient data-driven machine learning approach capable of instantly predicting the metal node types (paddle-wheel, rod, or others) of Cu-carboxylate MOFs, achieving excellent prediction accuracy (91%), precision (89%), and recall rates (85%), with the extreme gradient boosting (XGBoost) algorithm. Capitalizing on this powerful predictive model, we designed two tritopic carboxylate ligands functionalized with varied sterically hindered moieties and demonstrated the successful prediction of Cu-SBUs types through experimental implementation. The two newly synthesized Cu-MOFs (SJTU-2 and SJTU-5) consist of paddle-wheel or rod-like SBUs, showing novel interdigitated three-dimensional (3D) architecture formed through inclined interpenetration of 2D layers or 3D non-catenated <b>ths</b> net topology, respectively. SJTU-5 is a permanent ultramicroporous material featuring helical 1D channels with highly inert surfaces, rooted in abundant <i>tert</i>-butyl groups, which results in preferential adsorption of propane (C<sub>3</sub>H<sub>8</sub>) over propene (C<sub>3</sub>H<sub>6</sub>) with high ideal adsorbed solution theory (IAST) selectivity of 1.79 at ambient conditions. Overall, our work provides a reliable and reconfigurable machine learning model that exhibits exceptional potential in predicting the metal node types of MOFs while requiring no coding expertise, and would greatly facilitate the ultimate structural prediction and drive innovation in framework material development.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data-driven machine learning assisted prediction of metal node types in metal-organic frameworks for guiding linker design and targeting inverse C3H8/C3H6 separation

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Yifei Gao,
  • Pengfu Gao,
  • Ji Guo,
  • Yi Xie,
  • Jinqiao Dong,
  • Wei Gong,
  • Yong Cui

摘要

The structure prediction of metal-organic frameworks (MOFs) prior to synthesis is crucial for realizing rational discovery of materials with customized properties, while the extensive trial-and-error experimentation could be substantially reduced. However, the diversity of metal-containing secondary building units (SBUs) poses a major challenge to predicting the underlying net architectures from scratch. In this work, we developed an efficient data-driven machine learning approach capable of instantly predicting the metal node types (paddle-wheel, rod, or others) of Cu-carboxylate MOFs, achieving excellent prediction accuracy (91%), precision (89%), and recall rates (85%), with the extreme gradient boosting (XGBoost) algorithm. Capitalizing on this powerful predictive model, we designed two tritopic carboxylate ligands functionalized with varied sterically hindered moieties and demonstrated the successful prediction of Cu-SBUs types through experimental implementation. The two newly synthesized Cu-MOFs (SJTU-2 and SJTU-5) consist of paddle-wheel or rod-like SBUs, showing novel interdigitated three-dimensional (3D) architecture formed through inclined interpenetration of 2D layers or 3D non-catenated ths net topology, respectively. SJTU-5 is a permanent ultramicroporous material featuring helical 1D channels with highly inert surfaces, rooted in abundant tert-butyl groups, which results in preferential adsorption of propane (C3H8) over propene (C3H6) with high ideal adsorbed solution theory (IAST) selectivity of 1.79 at ambient conditions. Overall, our work provides a reliable and reconfigurable machine learning model that exhibits exceptional potential in predicting the metal node types of MOFs while requiring no coding expertise, and would greatly facilitate the ultimate structural prediction and drive innovation in framework material development.