Drug ranking using machine learning systematically predicts the efficacy of anti-cancer drugs

Area of Science:

• Computational biology
• Oncology
• Pharmacology

Background:

• Artificial intelligence and machine learning (ML) offer potential for personalized cancer therapy.
• Predicting drug efficacy requires robust analysis of complex biological data.

Purpose of the Study:

• To develop and validate a novel ML approach, Drug Ranking Using ML (DRUML), for predicting anti-cancer drug efficacy.
• To generate ordered lists of drugs based on their anti-proliferative effects in cancer cells using omics data.

Main Methods:

• DRUML utilizes internally normalized distance metrics of drug response as features for ML model generation, reducing noise and enhancing predictive robustness.
• The model was trained on in-house proteomics and phosphoproteomics data from 48 cancer cell lines.
• Validation was performed using data from 53 independent cellular models across 12 laboratories.

Main Results:

• DRUML accurately predicted drug responses in independent datasets with low error (MSE < 0.1) and high correlation (Spearman's rank = 0.7).
• DRUML's predictions of cytarabine sensitivity in leukemia patients were prognostic of patient survival (Log rank p < 0.005).

Conclusions:

• DRUML provides a robust and accurate method for ranking anti-cancer drugs based on predicted efficacy.
• The approach demonstrates potential for improving personalized cancer treatment strategies by identifying optimal therapies for individual patients.