AI, data, and the acceleration of drug discovery

Treatments for some of the world’s most prevalent and devastating diseases have come an extraordinarily long way in recent years.

For people with diabetes, the first medical treatment was introduced in 1922 when 14-year-old Leonard Thompson received an injection of insulin after Canadian doctors, Frederick Banting and Charles Best, figured out how to remove the hormone from a dog’s pancreas.

While revolutionary at the time (back then type 1 diabetes was always fatal), fast-forward several decades, and people with diabetes and obesity now have access to a new class of drug, semaglutide (aka Ozempic, Wegovy), which has been so effective and popular that the companies making the drugs are often unable to keep up with demand.

Those living with schizophrenia also have new-found hope following the very recent FDA-approval of Cobenfy, which represents an entirely new class of more targeted medicines for the treatment of schizophrenia, along with other difficult-to-treat neuropsychiatric and neurological diseases.

Read more: How Australian scientists helped pave the way for a new class of medicines to treat schizophrenia

These major, life-changing advancements make up a snapshot of truly great drug discovery progress, of which a far deeper pharmacological understanding of G-protein-coupled receptors (GPCRs) – the largest drug target class due to their involvement in signalling pathways related to many diseases – has significantly contributed to.

In fact, drugs that target the GPCR family make up about 34% of all US Food and Drug Administration (FDA) approved drugs, including Cobenfy and semaglutide.

However, while we have come a long way since early treatments for mental health conditions and that first dose of insulin, there are still many gaps when it comes to understanding how these drug receptors work and, importantly, why they might function differently in some people.

For example, a 2021 clinical trial found that 86.2% of participants achieved clinically significant weight loss after taking semaglutide for 68 weeks. While these numbers are very impressive, it also means that 13.7% of people did not see clinically significant weight loss. This could be attributed to a number of factors – including differences in receptor function among the participants.

When it comes to understanding inadequate side-effect profiles and why people with the same disease have differential effects when prescribed with the same medication, there’s still a lot of work to be done.

For a team of researchers from the Monash Institute of Pharmaceutical Sciences (MIPS), whose work largely revolves around GPCR-related drug discovery, a newly formed partnership with the Beijing-based team from Microsoft Research AI for Science seeks to address these ongoing challenges.

Together, the MIPS and Microsoft Research team is combining its expertise across datasets, drug discovery and artificial intelligence (AI), with the aim to develop a sequence-based AI-model to predict clinical impact of receptor polymorphisms (the occurrence of two or more clearly different morphs or forms) on treatments that help return the blood sugar to the normal range (that is, diabetes drugs).

The MIPS team, which is being led by Dr Elva Zhao, Professor Denise Wootten and Professor Patrick Sexton, are driven by how AI can help enhance their research in this space.

“Drug researchers and the broader pharmaceutical industry have been using computer and mathematical modelling for decades to help identify and design new drugs, but the development of predictive AI to identify patterns, anticipate behaviours and forecast outcomes has taken this field of research to another level,” says Dr Zhao.

“A really good example of this is the way in which AI can identify how an individual will respond to a drug, and also how one drug might interact with another when someone has been prescribed at least two medications.

“All of this information is critical in the context of tailoring drug design and treatments based on individual characteristics, including response to medications.”

But while both AI and advancements in drug discovery have both accelerated substantially in recent years, neither can reach their full potential without working together.

“The success of an AI-driven drug discovery program is dependent on the availability of high-quality data,” says Dr Zhao.

“Our goal is to integrate data, biological and pharmacological expertise with AI to create a working model that helps achieve our research goals and, ultimately, get to a place where we can generate new, important, information on receptor behaviour and function to help inform human responses to medications.”

The MIPS team will also work with Microsoft Research to explore how AI could enhance their existing research programs, which are largely centred on the development of new and improved drugs for metabolic diseases, such as type 2 diabetes and obesity.

“The global prevalence of metabolic diseases continues to rise, and there’s an ongoing need to improve the latest treatments to ensure equal opportunity for everyone who needs them,” says Dr Zhao.

“We’re really excited to work with the Microsoft Research team to customise its AI model for protein science, uncover new information and, hopefully, apply this to the development of new, targeted drugs that will greatly improve the lives of people, families and communities around the world.”