Cancer Drug Sensitivity Prediction Based on Deep Transfer Learning.

Area of Science:

• Computational biology
• Pharmacogenomics
• Drug discovery

Background:

• Phenotypic screening is crucial for drug discovery, but integrating diverse drug sensitivity data is challenging due to distribution discrepancies.
• Existing computational methods struggle to utilize multi-source pharmacogenomics data effectively for drug susceptibility prediction.

Purpose of the Study:

• To develop a deep transfer learning model for accurate drug susceptibility prediction across heterogeneous databases.
• To address the challenge of cross-database distribution discrepancies in pharmacogenomics data analysis.
• To create a reliable computational tool for precision drug development and personalized medicine.

Main Methods:

• Integrated genomic characterization of cancer cell lines with compound chemical information.
• Utilized the Encyclopedia of Cancer Cell Lines (CCLE) and Genomics of Cancer Drug Sensitivity (GDSC) datasets.
• Employed a domain-adapted deep transfer learning approach to predict half-maximal inhibitory concentrations (IC50 values).

Main Results:

• Successfully predicted drug susceptibility (IC50 values) by integrating multi-source heterogeneous data.
• Validated the predictive accuracy of the proposed deep transfer learning model.
• Demonstrated the model's ability to overcome cross-database distribution challenges.

Conclusions:

• The developed deep transfer learning model effectively predicts drug susceptibility, facilitating precision drug development.
• This approach enables the optimization of therapeutic strategies for personalized medicine.
• The model provides technical support for high-throughput drug screening and novel drug target discovery.