The drug discovery process is actually a lot like dating. You have a wide pool of potential candidates and put a lot of time and energy into weeding out the duds. A handful might make it past the awkward first dates and look as if they are in it for the long haul. But even after months and years convinced that this is the real deal ... it can all still fall apart.
The laborious, inefficient, and incredibly costly path to drug discovery has hardly changed in decades. Pharmaceutical companies spend billions over many years developing molecules that may ultimately fail at the final hurdle. Ripe for reinvention, it is now looking towards artificial intelligence as a means of dramatically accelerating the time to market for effective medicines, meaning they will get to the patients who need them much faster – and be significantly cheaper.
“We have this massive graveyard of all these compounds that have not worked,” says Dr Sudipto Das, a lecturer in pharmacy and biomolecular sciences at the RCSI. “Typically, researchers identify a target, develop a drug against it, and then test it – a drug might work in early phases of clinical trials but then ultimately fail down the line.
“This funnelling approach, where only a few drugs finally reach the clinic, is changing. AI will help to circumvent that process at the beginning where you are not investing a huge amount of resources in testing compounds that will ultimately not be effective.”
And it’s already happening. In March a paper published in Nature Biotechnology offered an in-depth look at how Insilico Medicine used AI to discover, develop and test its lead drug candidate, a treatment for the aggressive lung disease idiopathic pulmonary fibrosis (IPF), a process that took just three years.
Insilico is calling its achievement a “milestone” for generative AI in drug development, and says the latest paper functions as a blueprint for an industry undergoing a paradigm shift. Indeed, a report by Morgan Stanley suggests there is a $50 billion market over the next decade for AI in drug development.
Accenture’s global head of life science supply chain and industry, Barry Heavey, says it can cost “between $4 billion and $6 billion” to develop drugs and it takes anything from 10 to 15 years to get a new drug to market. That’s all set to change with the advent of “generative biology”, he says.
“We will be able to use generative AI and machine learning to understand biology in a way that is quicker than the traditional scientific experiments – everything is going to get faster,” adds Heavey.
Patricia Maguire is a professor of biochemistry at University College Dublin and also leads the UCD Institute for Discovery. She notes that even identifying the right targets for potential drugs can take years.
“Finding the precise underlying cause of a particular condition can take 20 years or more and there are so many where we still don’t even know it – we still don’t know what goes wrong in certain cancers, for example,” says Maguire.
Drugs work by binding to receptors, large protein molecules located on the surface of our cells, she explains. How this takes place can vary from person to person, and myriad factors can influence the process. This is where AI can have a huge impact, she says.
“I truly believe that AI can not only accelerate the basic understanding of diseases but also the functions of these protein receptors on the surface of our cells, so that we understand better what to target with potential drugs,” says Maguire.
This aligns with the push towards precision medicine, where the right drug at the right dose is given to the right patient. To do this, we need to know exactly what those protein structures look like.
Maguire explains that scientists only knew the structure of about a thousand proteins until very recently. However, in the past couple of years, the Alpha Fold project from Google DeepMind used AI to predict the structure of more than 200 million proteins. This will make all the difference when it comes to designing drugs, she says, as it will mean they will be more effective at lower doses, avoiding toxicities (side effects) and enhancing tolerability and safety.
“If you know what the structures of those proteins are you can design better drugs,” Maguire says. “It’s like a lock and key, and you know exactly what the lock looks like so you can design the best key. It means making a drug that ticks all the boxes – lowest possible dose with maximum efficacy and safety.”
It may also mean creating treatments that target multiple aspects of the same condition, or drugs that actually treat and prevent the root cause of disease. And when those locks have been determined, machine learning also allows for rapid screening of potential keys.
“It means we can screen those keys much faster because realistically we can’t just try every single one of them,” Maguire says. “AI is about bringing together all the different sources of data and that’s something that could never be achieved with a standard statistical approach before. It gives us a shortlist of effective keys much quicker.”
Heavey echoes this, saying AI will help to unravel the “many mysteries” of human biology. “It stands to reason that if you can use all the scientific experiments from the last 50 years in biology to train large language models and more powerful computing, that you can get to the bottom of the mysteries in biology quicker. And if you can do that then you can understand what drugs or combination of drugs might treat it,” he notes.
He also points out that the drugs that are being developed nowadays are much more complex – cell and gene therapies, viral vectors, MRNA – and, accordingly, they are more expensive to produce.
“The longer it takes to get the drug to market the less time the company has for getting a return,” Heavey says. “Anything that can be done to make that process less costly and more efficient is key.”
Talk of AI stirs up a lot of big feelings, however, and Heavey admits that many people are already dismissive of its potential in drug discovery.
“You still have to do a lot of experiments and prove to regulators and get patients on board with the concept,” he says. “There is a slight risk that we go through a ‘hype cycle’ where everyone thinks it’s going to happen tomorrow and then when it doesn’t it loses funding and there is a slowdown if things don’t play out as quickly as people hoped.”
Hype or not, the big hitters in biopharma are now focusing on AI across the entire drug development life cycle. Amgen has developed a machine learning model that can predict a protein’s viscosity so that scientists know how easy it will be to inject, for example. Pfizer says AI may help it predict the types of queries regulators will have about its new medicines, making it better prepared and thus ensuring the approval process is smoother and faster. And Sanofi recently inked a €140 million deal with Aqemia, a French start-up wholly focused on drug discovery using AI.
The pharma industry is grasping the nettle on this and Ireland will not be found wanting, Maguire believes.
“I see a huge opportunity here for Ireland Inc,” she says. “We could be world leaders in this.”
She has established the AI Healthcare Hub, bringing together a community of academics from across the university and connecting them with government, industry and tech companies on a global scale.
“In this rapidly evolving field, collaboration is essential”, she says. “With the right resources and the right people around the table, we could achieve this within the next five years in Ireland.”
Heavey’s term for the right people? Unicorns – not a typo. “I call them unicorns – people who understand the questions that need to be asked with the background and knowledge of biology, but who also understand the potential and the limitations of the computing system and AI. We need people who can traverse those two fields and they will be in demand.”
Das is training those unicorns. He is the lead on a new master’s degree programme at the RCSI, the MSc in Technologies and Analytics in Precision Medicine, which combines precision medicine, computational biology and digital health with a view to readying participants for this paradigm shift in drug discovery. Emerging courses such as this will allow Irish graduates and workers to apply their tech know-how to healthcare, he says.
“We have all those skills but there needs to be some level of adaptability in terms of applying them to healthcare data,” Das explains. “It is already happening to a certain degree but it is time to take it to the next level.”