Pateras, E., Vizirianakis, I.S. orcid.org/0000-0003-1459-9774, Zhang, M. et al. (3 more authors) (2025) Decoding Anticancer Drug Response: Comparison of Data-Driven and Pathway-Guided Prediction Models. Future Pharmacology, 5 (4). 58. ISSN: 2673-9879
Abstract
Background/Objective: Predicting pharmacological response in cancer remains a key challenge in precision oncology due to intertumoral heterogeneity and the complexity of drug–gene interactions. While machine learning models using multi-omics data have shown promise in predicting pharmacological response, selecting the features with the highest predictive power critically affects model performance and biological interpretability. This study aims to compare computational and biologically informed gene selection strategies for predicting drug response in cancer cell lines and to propose a feature selection strategy that optimizes performance.
Methods: Using gene expression and drug response data, we trained models on both data-driven and biologically informed gene sets based on the drug target pathways to predict IC50 values for seven anticancer drugs. Several feature selection methods were tested on gene expression profiles of cancer cell lines, including Recursive Feature Elimination (RFE) with Support Vector Regression (SVR) against gene sets derived from drug-specific pathways in KEGG and CTD databases. The predictability was comparatively analyzed using both AUC and IC50 values and further assessed on proteomics data.
Results: RFE with SVR outperformed other computational methods, while pathway-based gene sets showed lower performance compared to data-driven methods. The integration of computational and biologically informed gene sets consistently improved prediction accuracy across several anticancer drugs, while the predictive value of the corresponding proteomic features was significantly lower compared with the mRNA profiles.
Conclusions: Integrating biological knowledge into feature selection enhances both the accuracy and interpretability of drug response prediction models. Integrative approaches offer a more robust and generalizable framework with potential applications in biomarker discovery, drug repurposing, and personalized treatment strategies.
Metadata
Item Type: | Article |
---|---|
Authors/Creators: |
|
Copyright, Publisher and Additional Information: | © 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/ licenses/by/4.0/). |
Keywords: | pharmacogenomics; drug response prediction; feature selection; precision medicine; machine learning; biologically informed modeling; cancer cell lines |
Dates: |
|
Institution: | The University of Leeds |
Academic Units: | The University of Leeds > Faculty of Engineering & Physical Sciences (Leeds) > School of Mathematics (Leeds) |
Date Deposited: | 07 Oct 2025 10:49 |
Last Modified: | 07 Oct 2025 10:49 |
Published Version: | https://www.mdpi.com/2673-9879/5/4/58 |
Status: | Published |
Publisher: | MDPI |
Identification Number: | 10.3390/futurepharmacol5040058 |
Related URLs: | |
Open Archives Initiative ID (OAI ID): | oai:eprints.whiterose.ac.uk:232584 |