Kernelized rank learning for personalized drug recommendation.

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Large-scale cancer cell line screenings with molecular profiles aim to understand drug response genetics.
• Clinical drug recommendation differs from lab screens due to limited patient data, expert judgment, and focus on optimal therapy selection.

Purpose of the Study:

• To address limitations of current regression models in predicting drug sensitivity in clinical settings.
• To develop a machine learning approach for personalized drug recommendation that accounts for clinical realities.

Main Methods:

• Introduced Kernelized Rank Learning (KRL), a novel machine learning approach.
• Framed personalized drug recommendation as a rank learning problem, predicting drug effects per cell line (patient).

Main Results:

• KRL outperforms state-of-the-art predictors in drug recommendation, particularly with sparse training data.
• The approach demonstrates generalization capabilities to patient data.
• KRL effectively ranks drugs by predicted effect, bypassing precise sensitivity prediction.

Conclusions:

• KRL offers a promising machine learning solution for personalized drug recommendation in oncology.
• The method has significant translational potential for clinical applications.
• The Python implementation of KRL is publicly available for research use.