Machine learning model for anti-cancer drug combinations: Analysis, prediction, and validation

Jing-Bo Zhou; Dongyang Tang; Lin He; Shiqi Lin; Josh Haipeng Lei; Heng Sun; Xiaoling Xu; Chu-Xia Deng

doi:10.1016/j.phrs.2023.106830

Machine learning model for anti-cancer drug combinations: Analysis, prediction, and validation

Pharmacol Res. 2023 Aug:194:106830. doi: 10.1016/j.phrs.2023.106830. Epub 2023 Jun 19.

Authors

Jing-Bo Zhou¹, Dongyang Tang¹, Lin He¹, Shiqi Lin¹, Josh Haipeng Lei¹, Heng Sun¹, Xiaoling Xu², Chu-Xia Deng³

Affiliations

¹ Cancer Center, Faculty of Health Sciences, University of Macau, Macau SAR, China; Centre for Precision Medicine Research and Training, Faculty of Health Sciences, University of Macau, Macau SAR, China.
² Cancer Center, Faculty of Health Sciences, University of Macau, Macau SAR, China; Centre for Precision Medicine Research and Training, Faculty of Health Sciences, University of Macau, Macau SAR, China; MOE Frontier Science Center for Precision Oncology, University of Macau, Macau SAR, China.
³ Cancer Center, Faculty of Health Sciences, University of Macau, Macau SAR, China; Centre for Precision Medicine Research and Training, Faculty of Health Sciences, University of Macau, Macau SAR, China; MOE Frontier Science Center for Precision Oncology, University of Macau, Macau SAR, China. Electronic address: cxdeng@um.edu.mo.

PMID: 37343647
DOI: 10.1016/j.phrs.2023.106830

Abstract

Drug combination therapy is a highly effective approach for enhancing the therapeutic efficacy of anti-cancer drugs and overcoming drug resistance. However, the innumerable possible drug combinations make it impractical to screen all synergistic drug pairs. Moreover, biological insights into synergistic drug pairs are still lacking. To address this challenge, we systematically analyzed drug combination datasets curated from multiple databases to identify drug pairs more likely to show synergy. We classified drug pairs based on their MoA and discovered that 110 MoA pairs were significantly enriched in synergy in at least one type of cancer. To improve the accuracy of predicting synergistic effects of drug pairs, we developed a suite of machine learning models that achieve better predictive performance. Unlike most previous methods that were rarely validated by wet-lab experiments, our models were validated using two-dimensional cell lines and three-dimensional tumor slice culture (3D-TSC) models, implying their practical utility. Our prediction and validation results indicated that the combination of the RTK inhibitors Lapatinib and Pazopanib exhibited a strong therapeutic effect in breast cancer by blocking the downstream PI3K/AKT/mTOR signaling pathway. Furthermore, we incorporated molecular features to identify potential biomarkers for synergistic drug pairs, and almost all potential biomarkers found connections between drug targets and corresponding molecular features using protein-protein interaction network. Overall, this study provides valuable insights to complement and guide rational efforts to develop drug combination treatments.

Keywords: Bortezomib (PubChem CID: 387447); Cepharanthine (PubChem CID: 10206); Docetaxel (Taxotere) (PubChem CID: 11970251); Drug combination therapy; Flunarizine dihydrochloride (PubChem CID: 5282407); Lapatinib Ditosylate (PubChem CID: 9941095); Machine learning; Pazopanib hydrochloride (PubChem CID: 11525740); Potential combination biomarkers; Tamoxifen citrate (PubChem CID: 2733525); Three-dimensional tumor slice culture; Vatalanib (PubChem CID: 151194); Vincristine (PubChem CID: 5978); XL-184 (Cabozantinib) (PubChem CID: 25102847).

Publication types

Review
Research Support, Non-U.S. Gov't

MeSH terms

Antineoplastic Agents* / pharmacology
Antineoplastic Agents* / therapeutic use
Breast Neoplasms* / drug therapy
Drug Delivery Systems
Female
Humans
Phosphatidylinositol 3-Kinases
Signal Transduction

Substances

Phosphatidylinositol 3-Kinases
Antineoplastic Agents