Exciting news from Simon Fraser University! Researchers have just rolled out a new AI framework that could update the whole drug development process and speed up how we discover new medications.
This study could be a game changer in healthcare, as it proposes a fresh approach to one of the biggest hurdles in the pharmaceutical industry—designing and producing effective drug molecules. You can check out the full details published on the arXiv preprint server.
While AI has been showing great potential in crafting complex molecules aimed at fighting diseases, the real-life creation of these ‘ideal’ molecules often falls flat—many can’t actually be manufactured in a laboratory.
But there’s hope! This new AI method may drastically cut down the lengthy timeline required to bring drugs for common illnesses—think cancer—into play.
Martin Ester, who teaches computing science at SFU, shares, “Creating a new drug is a lengthy and costly affair. It’s commonly stated that the journey from conception to market for a new drug can take about a decade and cost upwards of $1 billion. We aim to simplify this process to fast-track the discovery and availability of innovative medicines that can combat diseases more promptly.”
One of the major roadblocks in AI-driven drug design is how to devise a feasible way to synthesize these molecules—the ‘recipe’ for making them. Without this crucial step, even the most promising AI inventions often get shelved, wasting time and resources in the process.
As Tony Shen, a Ph.D. student at SFU and the main author of the research, puts it, “Combating diseases starts with pinpointing the proteins that cause them. We then use computer models to design molecules that will latch onto these proteins, effectively neutralizing their harmful effects. Think of it as making a key that fits just right into a lock.”
The innovative method d CGFlow showcases a dual-design tactic where AI can model the construction of a molecule while also visualizing its 3D shape.
This combination is critical for creating molecules that are both biologically effective and feasible to synthesize.
As Ester notes, “We’ve come up with a machine-learning approach that makes it almost certain that the molecules we design can be chemically synthesized in the real world. This aspect is essential for making our models applicable to real-life problems, and it’s thrilling to see this come to fruition.”
Instead of cranking out entire molecules in one shot, CGFlow builds them one layer at a time—much like sculpting a statue piece by piece. This method allows the AI to adjust its understanding with each addition, which results in designs that are both more accurate and efficient.
But the excitement doesn’t stop in the lab; several companies are now eyeing the CGFlow framework for groundbreaking early-stage cancer drug discovery. There’s newfound hope in Developing treatments for complex illnesses.
“Our next move is to collaborate with industry partners to bring our method into practice and to refine CGFlow. We’re eager to witness how this can be utilized in real-world scenarios,” Ester concludes.
The findings were showcased at the International Conference on Machine Learning 2025 in Vancouver.
For more details: Tony Shen et al, Compositional Flows for 3D Molecule and Synthesis Pathway Co-design, arXiv (2025). DOI: 10.48550/arxiv.2504.08051
Provided by Simon Fraser University
This story first appeared on Tech Xplore.
