AI is about to remake the pharmaceutical industry – POLITICO

The next blockbuster drug could be invented by artificial intelligence.

For years, billions of dollars have been poured into bringing AI into the drug development process. Finally, it seems the cash might be paying off. 

Medicines designed by AI for conditions including lymph cancers, inflammatory diseases and motor neurone disease are reaching trials in humans. For many, it’s just a matter of time before they’re sitting on the shelf in a pharmacy.

If successful, AI promises nothing less than a revolution for the pharmaceutical industry: It could dramatically reduce the time it takes to develop a new medicine, as well as help identify new drug molecules that have so far eluded scientists. Drugmakers would pocket billions but it would also mean patients could have access to more new, innovative drugs at a pace not seen before.

But years of waiting for a result have left some skeptical that AI will actually deliver. While investors have been thronging to the space, no doubt spurred on by promises of a potential $50 billion opportunity, that won’t continue if AI-developed drugs don’t come to market. 

And there’s still one significant hurdle to overcome: The availability of data.

An industry on a precipice 

The premise for using AI in drug discovery and development is pretty straightforward: Use algorithms to trawl through vast troves of data — including the structures of chemical compounds, animal studies and information from patients — to help identify what a future drug needs to target in the human body; which molecule would be best suited for this; and most enticingly, how to create new molecules altogether. 

“I absolutely do believe that all drugs will be designed this way in the future, because I do believe it’s a far more efficient way to design molecules,” said Andrew Hopkins, the founder of Exscientia, which was one of the first companies to blaze onto the space in 2012. The question is how fast will the industry adopt this, he said.

But with high profile failures, such as the near-collapse of Sensyne Health and IBM’s Watson not living up to its big ambitions, there’s a lingering sense that AI might not meet the hype when it comes to developing new drugs. 

Currently, all eyes are on whether AI designed medicines will be safe for people, will have the desired effect on the disease, and will be able to meet the rigorous regulatory standards to actually be approved for human use.

“In the next few years, we need to see also clinical success of AI-driven projects, otherwise there will be a problem with translation to efficacy and safety of methods,” said Andreas Bender, professor of molecular informatics at Cambridge University and co-founder of AI drug discovery companies Healx and Pharmenable.

Jim Weatherall, AstraZeneca’s vice president of data science, AI and R&D, says the challenge for the next few years is pull through. If patients are actually going to benefit from drugs developed by AI, they need to receive them — which means the drugs will need to pass through the same regulatory hoops as traditional medicines.

With multiple AI-designed drugs now being tested in humans, it could soon be crunch time. 

Exscientia was first out the blocks in 2020 with a drug it hoped could treat obsessive compulsive disorder. While that study was discontinued after failing to reach expected criteria, the company now has a cancer drug and one for inflammatory diseases in clinical trials.

And it’s not the only one. 

Schrödinger has a potential lymphoma drug in clinical trials, Insilico has a drug to treat idiopathic pulmonary fibrosis that’s expected to enter phase 2 trials this year, and Verge Genomics is trialing a novel therapeutic for amyotrophic lateral sclerosis (ALS). 

In some cases, such as with Verge’s ALS drug, the drug development process itself has been thrown on its head. Traditionally, drugs are tested in animals before moving to humans. Instead, Verge’s platform uses human data and human models in the discovery and development phase, a process that they believe provides more insightful findings than animal models.

But it’s not a given that the first AI-drug approval will open the proverbial floodgates.

In different disease areas, there are different targets, different chemistry, meaning things look “entirely different,” said Bender. If there’s success with one drug “there will probably be public hype [and] more money coming into the area, but that doesn’t mean that all future projects are more likely to be successful,” he said. 

What’s needed is a better understanding of which data is predictive and meaningful in which disease context, in order to know which tool will be useful, in which situation, said Bender.

Critical ingredient

For a drug to even make it to clinical trials, the AI systems need be able to design it. To do that, they need access to immense quantities of data. That includes everything from data on the chemical composition of different molecules, to research papers and patient data. Without access to good quality and extensive data, the AI systems won’t be providing the most accurate results.

For a smaller company like the Italian biotech Dompé, that can be a major hurdle. 

“The new gap for me and a vision right now is the generation of high quality data in an amount that unlocks the true potential of artificial intelligence deep learning. These techniques do require a massive amount of validated data,” said Andrea Beccari, who heads up Dompé’s drug discovery platform. 

A company like Dompé won’t be able to do work at the scale of a large U.S. pharmaceutical company but if there were a central repository Beccari believes this could be a gamechanger for Europe — something like the European Health Data Space (EHDS) currently being negotiated by the Parliament and Council. 

The EHDS proposal seeks to make access to data for research much easier through the implementation of a system whereby researchers can request particular datasets through a permit. This could be data held by a public authority or even a pharmaceutical company, helping to level the playing field.

The EHDS promises to help create standards, improve interoperability and allow access to endless datasets. AstraZeneca’s Weatherall acknowledges that the opportunity is “tremendous”  but says that the key is doing this without “overbearing bureaucracy.” 

While AI promises to revolutionize the industry, those working on it for decades don’t see it as an algorithm doing all the work. Rather, it’s about “human in the loop AI,” said Weatherall. Yes, it’s a new way of working, but it just includes a few more experts in the room.

No going back

Those experts are increasingly being tapped on by Big Pharma. Industry is hedging its bets to ensure it’s not left in the wake of biotech startups that are flooding the space. Consultancy McKinsey estimates there are nearly 270 companies working in AI-driven drug discovery. While the majority are in the U.S., there are hubs emerging in Western Europe and Southeast Asia. 

In 2022, Pfizer extended its collaboration with an Israeli AI company; AstraZeneca expanded its cooperation with Benevolent AI; and Sanofi announced new work with Exscientia as well as a deal with Insilico Medicine.

In addition to partnerships like the one with BenevolentAI, British-Swedish drugmaker AstraZeneca has an in-house team of experts that are applying AI extensively in the drug discovery process. 

AI tools are applied to around 70 percent of the company’s drug discovery projects focused on small molecules — traditional drugs made from chemical compounds, said Ola Engkvist, associate director of computational chemistry, discovery sciences and R&D at AstraZeneca. They are also starting to use it outside of small molecules, in more complex projects such as antibody design.

There have already been successes in several projects, said Engkvist. While AI hasn’t yet created a new drug from start to finish, “maybe we are heading” there, he said.

Engkvist’s AstraZeneca colleague Weatherall believes that AI “is the future of drug development.” 

“We’ve been on a journey from ‘what is this?’ to ‘why did we ever do it any other way?’” he said.