Comparative evaluation of feature reduction methods for drug response prediction @SciReports 1. This study is the first to compare nine feature reduction (FR) methods for drug response prediction (DRP) on cell line and tumor transcriptomes, using over 6,000 machine learning (ML) model runs for robust analysis. 2. A key finding is that Transcription Factor (TF) activities outperform other FR methods on tumor data, distinguishing sensitive from resistant tumors for seven out of 20 drugs tested. 3. The researchers analyzed both knowledge-based methods (e.g., pathway and TF activities) and data-driven approaches (e.g., principal components), providing a broad perspective on FR for DRP. 4. Ridge regression emerged as the most effective ML model across all FR methods, highlighting its ability to handle correlated gene expression data. 5. Cross-validation on cell lines showed that sparse principal components and drug pathway genes performed best, while tumor validation emphasized the robustness of TF activities. 6. The study revealed that effective FR methods often depend on the drug and dataset type, underlining the need for tailored approaches in DRP. 7. TF activities proved to be a compact and interpretable representation of functional cellular states, bridging biological relevance and predictive accuracy in tumor DRP. 8. The findings emphasize the importance of robust FR methods to address the dimensionality challenges in molecular profiling, paving the way for precision medicine. @bennos @janbaumbach @behnam_bme @Faren_FIR 💻Code: github.com/faren-f/FS4DRP/tr… 📜Paper: nature.com/articles/s41598-0… #DrugResponsePrediction #MachineLearning #FeatureReduction #CancerResearch #PrecisionMedicine