Drug ranking using machine learning

Selecting the best drug for individual cancer patients is complicated. Cancer tumours differ dramatically from patient to patient, meaning that two people diagnosed with the same cancer may respond completely differently to the same therapy. It also makes it much harder to develop effective treatments; a consistent failure to accurately define the responsive patient cohort has led to a higher attrition rate in drug development.

By using machine learning to rank drugs, we could revolutionise cancer treatment through offering truly personalised medicine. In addition, machine learning could streamline drug development by providing a new and better way to stratify patients at scale – thereby making it easier to select the responding patient cohort for a cancer clinical trial.

Invented by Dr Pedro Cutillas, DRUML is a methodology for building and integrating machine learning models, using ensembles of proteomic, phosphoproteomic and transcriptomic features to generate lists of ranked drugs based on their efficacy.

Alongside its clinical potential, the technology can also be used as a research tool to narrow down who should participate in a drug trial by predicting whether individual patients are likely to be responsive.

DRUML can make predictions without needing to compare to samples – a crucial requirement for the clinical implementation of machine learning and a core aim of precision medicine.

Using large-scale proteomics and phosphoproteomics in machine learning has never been systematically applied before. However, recent advances in proteomic techniques and a greater number of drug response profiles means we can now feed any type of omics data into machine learning models of drug response – advancing the field of precision medicine and bringing hope to the millions of people diagnosed with cancer around the world each year.

A patent has been filed and we’re actively looking for partners to license this technology to develop commercially.