EUVIMEDOpdag EUVIMED
Crossref

Guided Ensemble Stacking Method for Predicting Biological Activities of Compounds

Azar Shamloo iD, Jack Tuszynski, Yun Tam, Chih-Yuan Tseng

Chemical Biology & Drug Design · 2025

DOI: 10.1111/cbdd.70230 ↗

At a glance

Journal article in Chemical Biology & Drug Design, 2025. Written by Azar Shamloo and 3 co-authors. Institution: Department of Physics, University of Alberta, Edmonton, Alberta, Canada. Major MEDLINE topics: Machine Learning. Cited 1 times so far (Crossref); 62 references. Funded by: Mitacs.

AbstractSubjectsBibliografiske oplysningerCiteRelated studies

Abstract (original language)

Machine learning (ML)‐driven quantitative structure–activity relationship (QSAR) modeling has gained significant attention for predicting compound biological activity based on compounds' structural, chemical, and physical properties because of the advancement of ML techniques. However, traditional ML‐QSAR models often suffer from biases due to algorithm selection and limitations in training data. Additionally, these approaches root in deducing biological activities purely from compounds' structural information and disregard their pharmacokinetic (PK) properties, a key factor contributing to the 15% failure rate in clinical trials, limiting their applicability in drug discovery. To address these challenges, we propose a guided ensemble‐based ML approach that integrates a supervised data preparation strategy with an ensemble stacking method, leveraging the strengths of multiple ML algorithms. By incorporating PK properties, our approach enhances prediction reliability. Specifically, we developed two ensemble stacking models: The classification model predicts the biological activity type, “inhibition” versus “activation,” based on compound features, while the regression model predicts bioactivity values. The classification model achieved an accuracy exceeding 0.85, while the regression model attained an R 2 above 0.77, demonstrating superior performance over traditional QSAR models. These results highlight the potential of our approach in improving drug discovery pipelines by enhancing predictive accuracy and addressing key QSAR limitations.

Subjects

MeSH terms (MEDLINE)

Machine LearningQuantitative Structure-Activity RelationshipAlgorithmsHumansDrug Discovery

Author keywords

PK propertiesQSAR modelsbiological activitiesstacking method

Authors and institutions

  • Department of Physics, University of Alberta, Edmonton, Alberta, Canada.
  • Department of Mechanical and Aerospace Engineering, Polytechnic di Torino, Turin, Italy.
  • SinoVeda Canada Inc., Edmonton, Alberta, Canada.

ORCID profiles: Azar Shamloo

Funding

  • Mitacs

Cite

APA

Shamloo, A., Tuszynski, J., Tam, Y., & Tseng, C. Y. (2025). Guided Ensemble Stacking Method for Predicting Biological Activities of Compounds. Chemical Biology & Drug Design. https://doi.org/10.1111/cbdd.70230

Vancouver

Shamloo A, Tuszynski J, Tam Y, Tseng CY. Guided Ensemble Stacking Method for Predicting Biological Activities of Compounds. Chemical Biology & Drug Design. 2025. doi:10.1111/cbdd.70230

BibTeX
@article{Shamloo2025,
  author = {Shamloo, Azar and Tuszynski, Jack and Tam, Yun and Tseng, Chih-Yuan},
  title = {Guided Ensemble Stacking Method for Predicting Biological Activities of Compounds},
  journal = {Chemical Biology & Drug Design},
  year = {2025},
  doi = {10.1111/cbdd.70230}
}
Start for free →