Predicting tumor cell line response to drug pairs with deep learning

Area of Science:

• Computational biology
• Pharmacogenomics
• Artificial intelligence in medicine

Background:

• The National Cancer Institute-60 (NCI-60) cell line panel provides a valuable resource for studying drug interactions.
• Understanding combinational drug activity is crucial for developing effective cancer therapies.

Purpose of the Study:

• To develop a computational model for predicting drug pair response in cancer cell lines.
• To leverage deep learning for modeling combinational drug effects.

Main Methods:

• Utilized residual neural networks to encode features and predict tumor growth.
• Integrated molecular data (gene expression, microRNA, proteome) and drug descriptors.
• Applied the model to the NCI-ALMANAC database.

Main Results:

• The model explained 94% of the response variance in cell line drug pair screening.
• Drug descriptors were found to be the primary drivers of predictive power.
• Successfully identified 80% of top-performing drug pairs with enhanced anticancer activity.

Conclusions:

• Deep learning shows promise for predicting combinational drug response.
• Future models could benefit from larger screening datasets to better utilize molecular features.