Latest News News

The Impact of Machine Learning on Drug Discovery

Researchers at McMaster University are among the first to employ machine learning, a subset of artificial intelligence, to predict effective compound combinations to treat drug-resistant fungal infections.

In a proof-of-concept study published in the journal Cell Systems, researchers from the Michael G. DeGroote Institute for Infectious Disease Research (IIDR), along with colleagues from the Universities of Montréal, Toronto and Edinburgh, describe a large scale chemical-genetic undertaking, where computational algorithms were used to identify synergistic pairs of small drug-like molecules.

Many areas of research are now using machine learning to find patterns in complicated datasets, says Jan Wildenhain, a co-lead author of the study who conducted the research while a graduate student in Dr. Michael Tyers’s Lab. “This trend has recently exploded in the biosciences because the amount of biological data available has simply become too large and complex to be processed by human intuition alone.”

Harnessing the predictive power of machine learning involves the development of specific algorithms, or queries, that can speed up the analysis of data. The researchers’ first attempt at creating an algorithm capable of identifying synergistic pairs of drug-like molecules was based on decades of data derived from Brewer’s yeast (S. cerevisiae) – one of the most powerful genetic models available.

But this initial algorithm, which included data on 200 diverse strains of yeast and their chemical-genetic responses to approximately 5,000 unique drug-like compounds, demonstrated weak predictive power, the authors write.

“This was a huge initial disappointment that sent us back to the drawing board,” says Micheala Spitzer, a postdoctoral fellow in the Wright Lab and co-lead author of the study. “We knew that chemical structures and the genetic network of cells had to be related to chemical synergisms we detected experimentally, but how to deconvolve these relationships from hundreds of thousands of data points was not obvious.”

A revised algorithm eventually yielded high predictive capabilities, leading the researchers to discover and validate more than 15 novel synergistic compound combinations.

This finding, says Wildenhain, is a vital step in expanding the current arsenal of antifungal drugs.

“Drug-resistant fungal infections are becoming increasingly prevalent, so there is a dire need for better antifungal drugs. Each year over a billion people, many of whom are immunocompromised, are affected by these types of infections.”

All of the study data is available for unrestricted download and exploration on a database called ChemGRID.