EUVIMEDEUVIMED entdecken
Crossref

Direct Prediction of Chemical Hazards in GHS Classification Using SMILES Representation for Health and Safety Applications

Takaaki Harada, Rumiko Hayashi, Kengo Tomita

ACS Chemical Health & Safety · 28.07.2025 · 32 (4) · 440-448

DOI: 10.1021/acs.chas.5c00035 ↗

Kurz gefasst

Fachartikel in der Zeitschrift ACS Chemical Health & Safety, online erschienen am 20.06.2025. Verfasst von Takaaki Harada und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 36 Literaturangaben. Gefördert von: Japan Society for the Promotion of Science.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Physical, health, and environmental hazards of chemical substances are classified into hazard classes in the Globally Harmonized System of Classification and Labeling of Chemicals (GHS). Within each hazard class in the GHS classification, the degree of chemical hazard is mostly expressed as a category. The chemical hazard information is in high demand for risk assessment and accident prevention; however, the hazards of most chemical substances are undetermined yet. Investigations on chemical hazards involve a time-consuming step such as in vitro and in vivo studies, whereas more and more novel chemical substances are synthesized on a daily basis. It is essential to accelerate investigations into the identification of chemical hazards. Herein, we use deep learning models to predict the hazard classes and categories of chemical substances in the GHS classification. The chemical structure is expressed as the Simplified Molecular Input Line Entry System (SMILES) notation. A hazard class prediction model is trained on a data set of chemical structures in SMILES that are labeled as to whether they are classified into the hazard class or not. Similarly, a hazard category prediction model is trained on a data set of SMILES labeled with a hazard category within the hazard class. The average accuracy values of hazard class and category prediction models are 89.2 and 70.4%, respectively. Test-time augmentation is employed to gain the ability of robust prediction regardless of the input chemical structure in SMILES. The atoms of a molecule that contribute to the correct prediction are determined by calculating their attribution scores, which provide insight into our prediction models. Our chemical hazard prediction models can be used as a tool for risk assessment and prioritization of physicochemical studies on novel chemical substances.

Förderung

  • Japan Society for the Promotion of Science

Zitieren

APA

Harada, T., Hayashi, R., & Tomita, K. (2025). Direct Prediction of Chemical Hazards in GHS Classification Using SMILES Representation for Health and Safety Applications. ACS Chemical Health & Safety, 32(4), 440-448. https://doi.org/10.1021/acs.chas.5c00035

Vancouver

Harada T, Hayashi R, Tomita K. Direct Prediction of Chemical Hazards in GHS Classification Using SMILES Representation for Health and Safety Applications. ACS Chemical Health & Safety. 2025;32(4):440-448. doi:10.1021/acs.chas.5c00035

BibTeX
@article{Harada2025,
  author = {Harada, Takaaki and Hayashi, Rumiko and Tomita, Kengo},
  title = {Direct Prediction of Chemical Hazards in GHS Classification Using SMILES Representation for Health and Safety Applications},
  journal = {ACS Chemical Health & Safety},
  year = {2025},
  volume = {32},
  number = {4},
  pages = {440--448},
  doi = {10.1021/acs.chas.5c00035}
}
Kostenlos starten →