Predicting Aqueous-Phase Hydroxyl Radical Reaction Rate Constants Using Machine Learning

Aqueous-phase hydroxyl radical (•OH) reaction rate constants control the environmental fate of organic compounds, yet experimental measurements remain unavailable for many compounds. Using 1229 organic species, this work established a machine learning framework that evaluates nine combinations of three molecular representations and three algorithms. All model selections used only five-fold cross-validation on the training set and out-of-fold (OOF) predictions; a held-out dataset was reserved for final evaluation. The molecular access system with random forest (MACCS–RF) combined model showed the best overall performance on the held-out test set with a coefficient of determination (R2) of 0.7001, a root mean square error (RMSE) of 0.0815, and a mean absolute error of 0.0567. OOF analysis identified several structural classes with relatively large prediction errors. The local model for sulfur-containing compounds slightly improved the overall results, increasing R2 to 0.7115 and reducing RMSE to 0.0799. SHapley Additive exPlanations (SHAP) analysis revealed that aromatic, cyclic, C=C-related, and sulfur-containing structural motifs contribute to •OH reactivity, while halogen and O–C–O substructures yielded negative contributions. Assessments of application domain (AD) using Tanimoto similarity and Williams plots indicated MACCS keys delivered much wider reliable coverage than Morgan fingerprints at the same thresholds. This study reveals structure-specific prediction biases for aqueous •OH reactivity and shows local modeling provides limited subclass-dependent benefits. By comparing chemical space coverage of different molecular representations, we propose an integrated workflow combining error diagnosis, structural interpretation, and AD evaluation for environmental kinetic quantitative structure–activity relationship (QSAR) modeling.

Authors

Institutions

Publication Details

Journal
Toxics
Published
2026-11-18
DOI
https://doi.org/10.3390/toxics14111017
Primary Topic
Computational Drug Discovery Methods
Type
article
Field-Weighted Citation Impact
0.00

Funders

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

Predicting Aqueous-Phase Hydroxyl Radical Reaction Rate Constants Using Machine Learning

Jingyi Yang, Yun Juan Wu, Xinlei Ge, Ziyan Zou et al.
1 links
Toxics
Computational Drug Discovery Methods
article

Predicting Aqueous-Phase Hydroxyl Radical Reaction Rate Constants Using Machine Learning

Jingyi Yang, Yun Juan Wu, Xinlei Ge, Ziyan Zou, Ning Ruan, Yunjiang Zhang, Tongtong Shou
article en

Abstract

Aqueous-phase hydroxyl radical (•OH) reaction rate constants control the environmental fate of organic compounds, yet experimental measurements remain unavailable for many compounds. Using 1229 organic species, this work established a machine learning framework that evaluates nine combinations of three molecular representations and three algorithms. All model selections used only five-fold cross-validation on the training set and out-of-fold (OOF) predictions; a held-out dataset was reserved for final evaluation. The molecular access system with random forest (MACCS–RF) combined model showed the best overall performance on the held-out test set with a coefficient of determination (R2) of 0.7001, a root mean square error (RMSE) of 0.0815, and a mean absolute error of 0.0567. OOF analysis identified several structural classes with relatively large prediction errors. The local model for sulfur-containing compounds slightly improved the overall results, increasing R2 to 0.7115 and reducing RMSE to 0.0799. SHapley Additive exPlanations (SHAP) analysis revealed that aromatic, cyclic, C=C-related, and sulfur-containing structural motifs contribute to •OH reactivity, while halogen and O–C–O substructures yielded negative contributions. Assessments of application domain (AD) using Tanimoto similarity and Williams plots indicated MACCS keys delivered much wider reliable coverage than Morgan fingerprints at the same thresholds. This study reveals structure-specific prediction biases for aqueous •OH reactivity and shows local modeling provides limited subclass-dependent benefits. By comparing chemical space coverage of different molecular representations, we propose an integrated workflow combining error diagnosis, structural interpretation, and AD evaluation for environmental kinetic quantitative structure–activity relationship (QSAR) modeling.

ToxicsVol. 14(11)
Nanjing University of Information Science and Technology (CN), Southeast University (CN)
National Natural Science Foundation of China
Computational Drug Discovery Methods
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