Machine learning-based drug response prediction in colorectal cancer with reduced features

Date

2026-7-29

Author

Coşgun, Mert

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

81
views

0
downloads

Cite This

The heterogeneous nature of colorectal cancer leads to significant variations in patients’ response to treatment processes. Even though various machine learning models have been implemented to predict the response to anticancer drug treatments, relying on complex and high-dimensional data limits their interpretability and real-world applications. In this thesis, various pipelines were developed with reduced features, integrating colorectal cancer cell lines’ gene expression and mutation profiles data with molecular compound fingerprints. The raw dataset consisted of 20,318 cell line–drug pairs, representing 47 colorectal cancer cell lines and 500 compounds. Drug response data, area under the dose-response curve (AUC) values, were discretized as sensitive and resistant. Overall, two discretization methodologies (Waterfall and Extreme Phenotype Sampling), four feature selection algorithms (Elastic Net Regression, LASSO, BORUTA, and mRMR), and three predictive models (Random Forest, XGBoost, and single-hidden layer neural network architectures) were implemented in all combinations. In the proposed pipeline using Extreme Phenotype Sampling for discretization and Elastic Net Regression for feature selection, an accuracy of 0.89 and an AUROC of 0.95 were achieved with both XGBoost and neural network architectures, requiring only 55 RNA-Seq and 26 mutation status features. The proposed pipeline framework offers a method for predicting colorectal cancer cell lines’ drug treatment outcomes. Additionally, this pipeline holds the potential to benefit clinical cancer treatment for colorectal cancer patients in prospective future applications that will require more rigorous futher validation steps.

Subject Keywords

Colorectal cancer, Precision oncology, Drug response prediction, Artificial intelligence, Computational biology

URI

https://hdl.handle.net/11511/120676

Collections

Graduate School of Natural and Applied Sciences, Thesis