A machine learning algorithm developed by researchers at the Queen Mary University of London in the UK can rank cancer drugs based on how effective they are at reducing the growth of cancer cells. The results were published in Nature Communications.
The researchers have called the algorithm 'Drug Ranking Using Machine Learning' (DRUML). Its models are based on datasets from analyses involving proteomics- the study of proteins inside cells- and phosphoproteomics- the study of how these proteins are modified- of 48 leukemia, esophagus, and liver cancer cell lines, and how they respond to 412 cancer drugs.
From this data, DRUML was able to accurately rank the different drugs in order of effectiveness at reducing cancer cell growth. DRUML was also able to predict cytarabine sensitivity, a kind of chemotherapy, in those with clinical leukemia, and predict patient survival.
The researchers cross-checked the algorithm's predictive accuracy with data from 12 other labs and a clinical dataset of 36 primary acute myeloid samples.
As new drugs are developed in the future, the researchers behind DRUML say their algorithm could be retrained to pinpoint the most effective drugs for various cancers at any given time.
"DRUML predicted drug efficacy in several cancer models and from data obtained from different laboratories and in a clinical dataset," said Pedro Cutillas, lead author of the study. "These are exciting results because previous machine learning methods have failed to accurately predict drug responses in verification datasets, and they demonstrate the robustness and wide applicability of our method."
Cancers vary greatly from patient to patient in terms of their genetic makeup and general characteristics, meaning that patients can respond very differently to different drugs. Machine learning methods such as DRUML could help doctors and researchers combine genetic markers with clinical and diagnostic information to identify optimal treatment options that are personalized to each patient.