Generously funded by Donors of Brain Tumour Foundation of Canada

Jonathan Stokes – McMaster University – Hamilton, ON

Project Title: “Generative AI for Novel drug design against recurrent glioblastoma”

Description of Project:

Adult brain tumors present a critical need for innovative therapeutic approaches, given the inadequacies of current treatments in achieving curative outcomes and providing limited survival benefits. About 1,000-1,500 Canadians are diagnosed with glioblastoma (GBM) every year, the most common and most lethal primary malignant brain tumor in adults. Despite aggressive multi-modal treatment, patients invariably relapse (recurrent GBM; rGBM) within only months post-diagnosis, resulting in a dismal prognosis of less than 15 months. Moreover, even after hundreds of clinical trials, there has been little progress over the past two decades.
Excitingly, contemporary artificial intelligence (AI) methods present a promising avenue to address the previous challenges associated with developing medicines to treat rGBM, due to their ability to explore truly vast chemical spaces in silico for novel molecules with the desired properties. In this project, we will leverage purpose-curated training datasets to train and deploy state-of-the-art generative AI algorithms that can design novel small molecule therapeutics against rGBM. Our novel AI-generated molecules will then be tested in the wet lab to validate their targeted anti-rGBM activity. This innovative project holds significant promise to improve quality of life for patients with rGBM and transform the paradigm of drug discovery in this challenging space.

What receiving this award means:

Through their remarkable generosity, the Brain Tumour Foundation of Canada has enabled our interdisciplinary research group to develop artificial intelligence technologies to help discover new therapies against recurrent glioblastoma. This work is challenging, but immeasurably high reward, and I am deeply grateful for the long-term vision of the BTFC in supporting this work at the edge of what is currently possible. I look forward to working with everyone at the BTFC to develop leading-edge artificial intelligence methods that can enable us to more rapidly invent new drugs to help patients.

Midpoint- August 2026

Glioblastoma is the most common and aggressive cancer that starts in the brain. Even with surgery, radiation, and chemotherapy, it almost always returns, and there are very few effective options once it does. Our project uses artificial intelligence to speed up the search for entirely new drugs to treat it. 

 We began in the laboratory by testing nearly 11,000 known biologically active molecules against glioblastoma cells grown from a patient’s tumour, identifying those that stopped the cancer from growing. We then used these results to teach an AI model the chemical “rules” that make a molecule effective against glioblastoma. Once trained, the model reviewed 12 million purchasable molecules in just four days – a task that would have taken 80 years by hand in the lab. 

 From the model’s predictions, we prioritized molecules that were likely to reach the brain, be safe, and be genuinely new, then purchased and tested 36 of them. Four of these 36 molecules killed glioblastoma cells effectively. Encouragingly, these four stayed active against cells taken from several different regions of the same tumour. This is important because glioblastoma varies enormously within a single patient. The most promising molecule, which we call E8619, was significantly more toxic to tumour cells than to normal cells, and it also blocked the cancer’s stem-like cells from regenerating the tumour, a key driver of recurrence. 

 However, E8619 had one serious flaw: it broke down within minutes in blood, far too quickly to work as an effective drug. We made a series of chemical variants of E8619 to fix this, but the very part of the molecule causing the metabolic instability turned out to be essential for its anti-cancer activity, so the two could not be separated. 

 Over the coming months, we are taking a more powerful approach. We are building a new generative AI method called a “chemical language model” that designs brand new molecules from scratch. Guided by our existing glioblastoma model, it will be steered toward compounds that are both potent against the tumour and free of the features that made E8619 unstable. We will draw from a pool of roughly three billion molecules that can be built and delivered in about a week, letting us design, make, and test new candidates rapidly and repeatedly. This rapid iteration will maximize the probability of discovering brand new molecules that meet all the criteria required to advance down the late-preclinical development pipeline.