Drug response profile-based machine learning enables strategic cell line and compound selection for drug development

Area of Science:

• Computational biology
• Pharmacogenomics
• Drug discovery

Background:

• Early drug discovery faces challenges in balancing biological diversity and experimental efficiency with limited cell line testing.
• High-dimensional omics data, while informative, can be costly and difficult to interpret for predictive modeling.
• Drug-response panel (DRP) descriptors offer a potential solution by capturing drug sensitivity profiles.

Purpose of the Study:

• To evaluate DRP descriptors as an efficient and informative alternative to omics data for modeling drug response.
• To assess the utility of DRP descriptors in cell line selection, biomarker discovery, and compound prioritization.

Main Methods:

• Utilized gradient boosting models on GDSC and CCLE datasets.
• Employed DRP descriptors and mRNA expression features to predict drug sensitivity (-log10(IC50)).
• Interpreted models to identify known signatures and potential biomarkers.

Main Results:

• DRP descriptors consistently outperformed mRNA expression features in predicting drug sensitivity.
• Model interpretation revealed MAPK-associated sensitivity signatures and potential biomarkers for MEK1/2 and BTK/MNK inhibitors.
• Successfully distinguished between tumorigenic and non-tumorigenic cell lines, enabling selective compound identification.

Conclusions:

• DRP-based representations derived from compact screening panels enable efficient cell line selection.
• This approach supports biomarker discovery and compound prioritization in early-stage drug development.
• DRP descriptors provide a valuable tool for enhancing the efficiency and informativeness of drug discovery processes.