High-Throughput Screening of Photocatalytic Metal–Organic Frameworks for CO2 Reduction Using Relational Data Extraction and Hierarchical Machine Learning

Photocatalytic carbon dioxide reduction (CO2RR) over metal–organic frameworks (MOFs) is governed by both framework features and reaction conditions, which complicate catalyst discovery and render general design rules difficult to establish. In this study, we develop a synergistic machine-learning workflow to integrate literature data extraction, electronic-property prediction, and hierarchical classification for prioritizing MOF photocatalysts. First, a literature-derived record-level data set containing catalyst compositions, structural descriptors, reaction conditions, and catalytic outcomes was curated through large language model (LLM)-assisted relation extraction followed by manual verification. Then, a transfer-learning crystal graph convolutional neural network (CGCNN) model was trained for band gap prediction at the Heyd-Scuseria-Ernzerhof screened hybrid functional (HSE06) level, achieving a mean absolute error (MAE) of 0.31 eV against reference HSE06 calculations. Subsequently, a physically constrained multitask framework was employed to predict electron affinity (EA) and ionization potential (IP) for thermodynamic prescreening. Three record-level classifiers were constructed for activity screening (C1), performance classification (C2), and product selectivity prediction (C3). When applied to the Computation-Ready, Experimental (CoRE) MOF database under standardized gas- and liquid-phase inference settings, the workflow prioritized 639 primary candidates, with most assigned to CO formation. Probability calibration and applicability-domain analysis further identified 166 of these candidates (25.98%) as a more stringent calibrated, in-domain subset. Overall, this study provides a practical data-driven workflow for condition-resolved prioritization of MOF photocatalysts and analysis of descriptor-level trends associated with photocatalytic CO2RR.

Authors

Institutions

Publication Details

Journal
Journal of the American Chemical Society
Published
2026-11-18
DOI
https://doi.org/10.1021/jacs.6c14994
Primary Topic
Machine Learning in Materials Science
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

High-Throughput Screening of Photocatalytic Metal–Organic Frameworks for CO2 Reduction Using Relational Data Extraction and Hierarchical Machine Learning

Yong Cui, Xiao Niu, Jianwen Jiang, Yan Liu
3 links
Journal of the American Chemical Society
Machine Learning in Materials Science
article

High-Throughput Screening of Photocatalytic Metal–Organic Frameworks for CO2 Reduction Using Relational Data Extraction and Hierarchical Machine Learning

Yong Cui, Xiao Niu, Jianwen Jiang, Yan Liu
article en

Abstract

Photocatalytic carbon dioxide reduction (CO2RR) over metal–organic frameworks (MOFs) is governed by both framework features and reaction conditions, which complicate catalyst discovery and render general design rules difficult to establish. In this study, we develop a synergistic machine-learning workflow to integrate literature data extraction, electronic-property prediction, and hierarchical classification for prioritizing MOF photocatalysts. First, a literature-derived record-level data set containing catalyst compositions, structural descriptors, reaction conditions, and catalytic outcomes was curated through large language model (LLM)-assisted relation extraction followed by manual verification. Then, a transfer-learning crystal graph convolutional neural network (CGCNN) model was trained for band gap prediction at the Heyd-Scuseria-Ernzerhof screened hybrid functional (HSE06) level, achieving a mean absolute error (MAE) of 0.31 eV against reference HSE06 calculations. Subsequently, a physically constrained multitask framework was employed to predict electron affinity (EA) and ionization potential (IP) for thermodynamic prescreening. Three record-level classifiers were constructed for activity screening (C1), performance classification (C2), and product selectivity prediction (C3). When applied to the Computation-Ready, Experimental (CoRE) MOF database under standardized gas- and liquid-phase inference settings, the workflow prioritized 639 primary candidates, with most assigned to CO formation. Probability calibration and applicability-domain analysis further identified 166 of these candidates (25.98%) as a more stringent calibrated, in-domain subset. Overall, this study provides a practical data-driven workflow for condition-resolved prioritization of MOF photocatalysts and analysis of descriptor-level trends associated with photocatalytic CO2RR.

Journal of the American Chemical Society
National University of Singapore (SG), Shanghai Jiao Tong University (CN)
Shanghai Key Basic Research Program, National Research Foundation Singapore, National Natural Science Foundation of China, China Scholarship Council, National Key Research and Development Program of China
Climate action
Machine Learning in Materials Science
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon