Pan-Cancer Prediction of Cell-Line Drug Sensitivity Using Network-Based Methods

Area of Science:

• Computational Biology
• Genomics
• Pharmacology

Background:

• Developing predictive models for anti-cancer drug response is vital for personalized medicine.
• Challenges include vast differences in cancer cell lines and drug characteristics, limiting model accuracy.
• Existing models struggle to explain similarities between cell lines or drugs.

Purpose of the Study:

• To propose a novel network-based methodology to improve the predictive power of anti-cancer drug responses in cell lines.
• To break down the complex prediction problem into smaller, interpretable sub-problems.
• To identify biological correlates of drug response.

Main Methods:

• Utilized the Genomics of Drug Sensitivity in Cancer (GDSC) database (915 cell lines, 200 drugs).
• Applied optimal mass transport theory for network-based clustering of cell lines (gene expression) and drugs (cheminformatics features).
• Employed random forest regression for predictive modeling within cell-line-drug clusters, followed by biological analysis.

Main Results:

• Developed 30 distinct cell-line-drug clusters using the network-based approach.
• Predictive models on clustered subsets significantly outperformed alternative computational methods.
• Identified three of the top four drugs by prediction performance targeting the PI3K/mTOR signaling pathway.

Conclusions:

• Network-based clustering improves prediction accuracy for cell-line specific anti-cancer drug responses.
• This methodology enhances the interpretability and predictive power of drug response models.
• Post-modeling analysis revealed plausible biological processes linked to drug responses.