MinDrug

MinDrug predicts anti-cancer drug responses (IC50) for novel cell lines by integrating drug similarity and connectivity information with Elastic-Net regression to support personalized oncology research.

Key Features:

• Optimal Drug Subset Identification: Uses a heuristic star algorithm to identify an optimal subset of drugs that exhibit high similarity to other drugs within large datasets.
• Predictive Modeling: Applies Elastic-Net regression to predict half-maximal inhibitory concentration (IC50) values for anti-cancer drugs on novel cell lines while integrating drug similarity and connectivity information.
• Validation and Comparative Evaluation: Employs statistical and biological validation methods and conducts comparative analyses against four existing approaches using k-fold cross-validation.
• Public and External Dataset Evaluation: Evaluated on the Genomics of Drug Sensitivity in Cancer (GDSC) and the Cancer Cell Line Encyclopedia (CCLE) and tested on an external dataset with a differing statistical distribution to assess generalizability.
• Performance Attributes: Demonstrates precision, robustness, and computational speed compared to existing methods.

Scientific Applications:

• Personalized treatment strategy development: Predicts how different cell lines respond to anti-cancer drugs to inform individualized therapy decisions.
• Preclinical evaluation and drug selection: Identifies relevant drugs and prioritizes candidates for further experimental study using large-scale cell line and drug datasets.

Methodology:

Heuristic star algorithm for optimal drug subset identification; Elastic-Net regression for IC50 prediction integrating drug similarity and connectivity information; statistical and biological validation including k-fold cross-validation and comparative analyses against four existing approaches; evaluation on GDSC, CCLE, and an external dataset with a differing statistical distribution.