Related Experiment Video
Updated: Aug 4, 2026
07:42
Assessment of Resistance to Tyrosine Kinase Inhibitors by an Interrogation of Signal Transduction Pathways by Antibody Arrays
Published on: September 19, 2018
8.0K
Machine learning classifier approaches for predicting response to RTK-type-III inhibitors demonstrate high accuracy
Mauricio H Ferrato1, Adam G Marsh1, Karl R Franke2
1University of Delaware, Newark, DE 19716, USA.
Bioinformatics Advances
|May 30, 2023
Summary
Machine learning models can predict patient response to cancer drugs using gene expression data. The Shapley Additive Explanation (SHAP) technique, combined with random forest, achieved 89% accuracy in predicting response to Foretinib. This highlights ML
Area of Science:
- Bioinformatics
- Computational Biology
- Precision Medicine
Background:
- Machine learning (ML) shows promise in precision medicine but faces challenges in predicting therapy response.
- Accurate classification of therapy responders versus non-responders remains an active research area.
- Developing novel ML approaches is crucial for advancing personalized treatment strategies.
Purpose of the Study:
- To investigate the impact of feature selection techniques and classifiers on ML model performance for predicting patient response to RTK-type-III inhibitors.
- To leverage gene expression data and ML to identify transcriptomic signatures predictive of treatment outcomes.
- To evaluate the efficacy of Principal Component Analysis (PCA), SHAP, and differential gene expression analysis in conjunction with XGBoost, LightGBM, and Random Forest (RF) classifiers.
Main Methods:
- Utilized publicly available RNA-seq gene count data from 451 individuals via the BeatAML initiative.
- Applied three feature selection methods: PCA, SHAP, and differential gene expression analysis.
- Tested three classifiers: XGBoost, LightGBM, and RF, evaluating model performance using Area Under the Curve (AUC)-Receiver Operating Characteristic (ROC) curves.
Main Results:
- Feature selection technique significantly impacted model performance more than the choice of classifier.
- The SHAP technique demonstrated superior performance compared to PCA and differential gene expression analysis.
- The highest performing model combined SHAP with RF classifier, achieving an 89% AUC for predicting response to Foretinib.
Conclusions:
- A transcriptomic signature exists at diagnosis that can potentially predict treatment response.
- ML applications hold significant potential for advancing precision medicine efforts.
- The developed ML pipelines offer a promising approach for personalized cancer therapy selection.
