Researchers from Simon Fraser University have unveiled an artificial intelligence framework that could transform drug development and accelerate the discovery of new medicines.
In a potential breakthrough for healthcare, the new study presents an innovative method that tackles one of the pharmaceutical industry's most persistent challenges - how to design and make effective drug molecules.
For years, AI tools have shown great promise in designing complex molecular structures that can, in theory, interact with disease targets. Yet, many of these 'perfect' molecules prove impossible to manufacture in real-world labs.
It is hoped that this new process could dramatically reduce the amount of time it takes to discover and manufacture drugs used to treat common diseases, such as cancer.
Martin Ester, professor of computing science at SFU, says: "The development of a new drug is an extremely time consuming and expensive process. As a rule of thumb, people always say that it takes 10 years and $1 billion USD to bring a new drug to market.
"Our hope is that our method will significantly shorten this process so that new drugs can be discovered, produced and made available in a much shorter timeframe in order to help cure diseases."
One of the core challenges for AI drug design is the synthesis pathway - the ability to come up with a realistic chemical recipe to build the molecule. Without this, even the most promising AI-designed molecules are often discarded, leading to wasted time and resources.
"The fight against disease starts with identifying the disease-causing protein," explains Tony Shen, SFU PhD student and lead author of the paper.
"In the lab, computer models are then used to design molecules that will bind to the disease-causing protein, often deactivating it and stopping its harmful activity. The whole process is a bit like trying to design a key that will fit into a lock."
The new method presented in the study, called CGFlow, introduces a dual-design approach that enables AI to simultaneously model how a molecule is constructed and what it looks like in 3D space.
This combination is essential for generating molecules that are not only biologically effective but also chemically feasible to produce.
"We have developed a machine-learning method that practically guarantees that the molecule generated can be created through chemical synthesis in the real world," says Ester.
"This is a hugely important aspect in translating the results of these generative models into practical applications, it is very exciting."
Instead of designing molecules in one go, CGFlow assembles them step by step, much like sculpting a statue by adding one piece of clay at a time.
With each step, the AI learns how the new component changes the overall shape and function of the molecule, resulting in more accurate and efficient designs.
The model's potential is already being recognized beyond the lab. Several companies are looking at adopting the CGFlow framework for early-stage cancer drug discovery, offering new hope in the fight against complex diseases.
"The next step is to take our method to industry so it can be used and improved. We're really interested in working with industry to evaluate and further develop CGFlow in practical applications," adds Ester.
The study was published at the International Conference on Machine Learning 2025, in Vancouver, a top conference in its field.
Available SFU Experts
TONY SHEN, PhD student, computing science | [email protected]
MARTIN ESTER, professor, computing science | [email protected]
