Canadian supercomputers are transforming drug discovery: Can they deliver medicines faster?

The traditional process of drug discovery is lengthy and complex. Scientists must identify disease-causing biological targets, design molecules capable of interacting with those targets, test thousands of candidates, and then determine whether any are safe and effective enough to progress into clinical development. It is an expensive undertaking that is frequently described as taking around a decade and billions of dollars to bring a successful medicine to market.

Artificial intelligence (AI) is increasingly being positioned as a solution to this challenge. Across academia and industry, researchers are combining AI algorithms with powerful computational infrastructure to identify promising drug candidates more rapidly. One notable example comes from Simon Fraser University (SFU) in Canada, where researchers have developed an AI framework aimed at accelerating the design of disease-fighting drug molecules while also addressing one of the biggest barriers in computational drug discovery: ensuring that the molecules can actually be made in a laboratory.

Drug discovery has always been heavily dependent on computation. Molecular modelling, virtual screening and structure-based drug design have been used for decades to help identify promising compounds. What AI brings is the ability to analyse vastly larger datasets and explore far larger areas of chemical space than conventional approaches.

Researchers at Simon Fraser University report that their new framework, called CGFlow, is designed to tackle a longstanding problem. While AI systems have become increasingly capable of designing molecules that theoretically bind to disease-causing proteins, many of these computer-generated molecules cannot be synthesised using real-world chemistry. Consequently, potentially promising computational discoveries often fail before laboratory testing even begins.

According to the SFU team, CGFlow simultaneously models both the three-dimensional structure of a molecule and the chemical pathway required to create it. This dual-design approach aims to generate molecules that are not only biologically active but also manufacturable.

The work highlights an important shift in AI drug discovery. Rather than simply generating theoretical molecular structures, modern systems increasingly seek to incorporate practical chemistry constraints into their designs.

The role of academic supercomputing

Artificial intelligence requires substantial computational resources. Drug discovery models may need to evaluate millions or even billions of molecular possibilities. This demands access to high-performance computing systems capable of executing vast numbers of calculations simultaneously. Academic supercomputing facilities increasingly provide the infrastructure necessary to support such research. By combining machine learning algorithms with advanced computing environments, researchers can perform molecular simulations, predict protein interactions, evaluate compound properties and rank potential drug candidates at speeds that would be impossible using traditional laboratory-only approaches.

In the SFU programme, researchers are collaborating with scientists associated with the Vancouver Prostate Centre, an organisation that already has substantial expertise in computational drug discovery, cheminformatics and translational medicine. The Centre’s drug discovery capabilities include computational screening, structure-activity modelling and computer-aided optimisation of candidate compounds.

The combination of AI, advanced computing and laboratory validation represents a powerful multidisciplinary approach that is becoming increasingly common throughout pharmaceutical research.

From proteins to potential medicines

At its core, drug discovery involves finding molecules capable of interacting with biological targets. The SFU researchers describe the process as similar to designing a key that fits a lock. The disease-causing protein represents the lock, while the therapeutic molecule acts as the key. If the molecular fit is sufficiently precise, the drug may inhibit, activate or otherwise alter biological activity. Historically, identifying such molecules has required extensive experimentation and large compound libraries.

AI fundamentally changes this process by enabling researchers to model molecular interactions in silico before compounds are synthesised and tested in the laboratory. This significantly narrows the field of candidates requiring experimental evaluation. The potential impact extends beyond cancer treatments. Researchers have suggested the technology may also be applied to areas including autoimmune disorders and inflammatory diseases.

For pharmaceutical companies, drug development remains one of the most expensive forms of research and development. Attrition rates are high, and many promising candidates fail during development. Hence, any technology capable of improving target identification, accelerating hit discovery or reducing laboratory testing requirements has the potential to lower development costs and shorten research timelines.

Internal pharmaceutical research has already been exploring AI applications ranging from compound screening and biomarker identification through to clinical trial optimisation. Enterprise research and training materials continue to identify drug discovery as one of the most promising applications of artificial intelligence within the pharmaceutical sector.

Other breakthroughs are also emerging. Systems such as DeepMind have demonstrated how AI can accelerate protein structure prediction, providing researchers with important insights into biological targets that were previously difficult or impossible to obtain efficiently. Enterprise training materials note that such advances may significantly improve drug target identification and pharmaceutical research capabilities.

Despite the excitement, significant hurdles remain. Identifying a promising molecule computationally does not guarantee success in the clinic. Drug candidates must still undergo synthesis, biological testing, toxicology studies and multiple phases of clinical trials. There are also concerns around model reliability, data quality and reproducibility. AI models are ultimately dependent upon the quality of the datasets used to train them.

Furthermore, pharmaceutical companies must operate within highly regulated environments. If AI-generated outputs influence pharmacological decisions, questions arise regarding model validation, transparency and documentation. This is particularly relevant because regulators increasingly expect traceability and scientific justification for key development decisions.