Drug Selection via Joint Push and Learning to Rank

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Precision medicine aims to match patients with optimal therapies.
• Cancer drug selection is complex, requiring prediction of drug efficacy for individual patients.
• Existing methods for predicting drug response in cancer cell lines have limitations.

Purpose of the Study:

• To develop a novel learning-to-rank method for accurate cancer drug selection.
• To predict the ranking of sensitive drugs and their order within cancer cell lines.
• To leverage genomics information for improved drug response prediction.

Main Methods:

• Formulated cancer drug selection as a learning-to-rank problem.
• Developed a new method, pLETORg, utilizing drug and cell line latent vectors.
• Incorporated genomics data to learn latent vectors and enforce correct drug ranking.

Main Results:

• pLETORg accurately predicts drug ranking positions and orders among sensitive drugs.
• The method effectively prioritizes sensitive drugs over insensitive ones.
• Experimental results show pLETORg significantly outperforms state-of-the-art methods.

Conclusions:

• pLETORg offers a significant advancement in cancer drug selection for precision medicine.
• The method demonstrates improved accuracy in identifying effective cancer drugs.
• Leveraging genomics data within a learning-to-rank framework enhances drug response prediction.