Drug development is often a long and tedious process; sometimes, taking over a decade and billions of dollars before a treatment reaches patients. In fact, over 90 per cent of drug trials fail in the early stages.
New research at Queen’s is aiming to change that using artificial intelligence (AI) to determine successful drug candidates, receiving the 2025 John Charles Polanyi Prize in Chemistry.
The Prize is awarded annually to five researchers who’re in the early stages of their career, continuing their postdoctoral studies or appointed to a faculty position at an Ontario university.
For 2025, the prizes have a value of $20,000 each and are broadly defined in the categories of Physics, Chemistry, Physiology or Medicine, Literature, and Economic Science.
In an interview with The Journal, Dr. Fawang Meng, the recipient of the 2025 Polanyi Prize in Chemistry and a Banting postdoctoral fellow in the Department of Chemistry, shared insight into his research.
“My research focuses on the early stages of drug discovery where we want to identify molecules or compounds with desired properties that make it a good candidate for a drug trial,” Meng said.
Using machine learning, Meng’s work helps identify which compounds satisfy multiple properties, such as effectiveness, safety, and chemical stability, before moving them forward in drug development.
According to Meng, datasets produced by advances in computing power and automated laboratory systems are key to training modern machine learning models. “We have accumulated an enormous amount of data points, and those data points provide the foundation to build up really powerful machine learning models,” Meng added.
By analyzing those datasets, AI learning models can predict how different chemical compounds might behave. “Based on those predictions, we can select a few of the most promising [compounds], and submit those compounds into experimental settings. The new data can also feed back into the model, so the model can use it for the next round of prediction,” Meng said.
Before machine learning became widely used, computational drug discovery relied on traditional simulation methods, which were not always reliable.
“The speed can be a really big issue. Sometimes, they’re really slow to get and it’s hard to say if your prediction is accurate or not,” Meng noted. “What helps AI models stand out is efficiency, speed, and accuracy.”
However, applying machine learning to chemistry poses its own challenges with respect to the quality of available data. “The data sets can contain a lot of missing values,” Meng said. “The other thing is the data set can be imbalanced, and some data sets are actually very noisy because of the experimental setting.”
“It’s normal to throw out data with missing values, which often leads to information loss. What I did was try to build a model which maximizes the value of each data point,” Meng added.
Improving how models handle imperfect datasets could lead to more reliable predictions and ultimately better drug candidates.
For now, Meng sees his work as part of a broader effort to strengthen the tools that scientists use to discover medicines.
Tags
chemistry department, drug trial, machine-learning, Polanyi Prize
All final editorial decisions are made by the Editor(s) in Chief and/or the Managing Editor. Authors should not be contacted, targeted, or harassed under any circumstances. If you have any grievances with this article, please direct your comments to journal_editors@ams.queensu.ca.