PharmaTimes speaks to AI expert Jackie Hunter about the future of these technologies in R&D

Jackie Hunter, CEO of BenevolentAI, thinks that lists of the top 20 pharma companies will look very different in five years time thanks to AI.

“If you look at the top 20 today compared to ten years ago there are lots of different players in there, such as Teva and Gillead, who weren’t there 20 years ago, and that’s been driven by the biotech revolution and new disease understanding,” she says, speaking at a breakfast briefing organised by industry recruitment specialists PiR. “So I don’t see why AI won’t drive a similar change. You might even see different players like Google enter the list. Google have already invested heavily in biotech and diagnostics, so they are very well placed to do that.”

AI is a rapidly growing trend in pharma, with several big pharma firms – including Merck & Co, Johnson & Johnson, Pfizer and Sanofi – now exploring the area. GSK, for example, recently signed a deal worth up to £33 million with Dundee-based AI company Exscientia to collaborate on drug discovery.

Even so, Hunter says that she knows of no pharma companies with an integrated AI strategy – even though she thinks it could help fix pharma’s “unsustainable” drug discovery model, which currently has a 50 percent failure rate in both phase 2 and phase 3.

In BenevolentAI’s case, the platform mostly works by analysing data and finding connections between it. When working in motor neurone disease/amyotrophic lateral sclerosis (ALS), for example, the AI the company developed was able to review billions of sentences and paragraphs of scientific research papers and abstracts to find relationships between the data which it regulated into ‘known facts’. These facts were curated and new connections were made to generate a large number of possible hypothesis using criteria set by the scientists – in the case of ALS, there were around 200.

The researchers could then assess the validity of the hypotheses and arrive at a prioritised list of those which were considered to be worth exploring. Eventually five hypotheses were chosen to be explored in the lab. The results were largely successful – of these drugs, one had effect in in-vivo models of ALS, two showed excellent efficacy in assays, two showed lesser but significant effect and one had no effect.

For this reason, Hunter does not believe that AI will lead to massive losses in clinical jobs, and says that the reality of the technology is actually closer to ‘augmented intelligence’ than artificial intelligence.

“Researchers still need to use their intelligence and experience to interrogate those hypotheses,” she says, “but they’re getting a much broader base of evidence with a reduction in bias.”

This is also an important component in being able to maintain a drug’s patentability:  “For patentability you have to have a scientist overlay their conditional interpretation into the data, because if everything is automatic you could say where the inventive step is?”

To be able to find these relationships, BenevolentAI has had to build a database of tens of millions of publications that the AI can read. “We take a huge variety of data sources – both structured data and unstructured data. The thing about this data is it can be messy, and we’ve spent two years building the domain-specific dictionaries that allow us to overcome that. For example, AD could refer to atopic dermatitis or Alzheimer’s disease, so the AI needs to understand the context in which AD occurs.”

Hunter says AI can also help overcome the problem of siloed data. “The corpus of information from oncology, for example, can inform antimicrobial resistance, or information from Alzheimer’s can inform atopic dermatitis, yet without using AI for a broader view it is usually organsied in silos.”

Realities of the industry

One initial problem companies might face is getting AI technicians who have never worked with pharma to work together with clinical researchers who have never worked with AI. This was true for BenevolentAI, but they did find a solution : “We incentivised them to work together by making sure their objectives were shared, and setting expectations that the culture of the organisation had to be different to facilitate communication,” says Hunter.

New methods of drug discovery involving AI may also result in a need for regulators to change how they assess drugs, but Hunter thinks that “regulations are way behind where they need to be in terms of an understanding of using these technologies”.

“Sometimes I’m surprised by how little some regulators actually know about the drug development process. Education for regulators about the process of R&D is essential, and this is where bodies like IMI, PhRMA and EFPIA need to come together to make the case for AI.

“The one message I would give to regulators is don’t impose regulations that stifle innovation, because ultimately patients will be the ones that won’t benefit. By overburdening the regulatory system you could end up hampering innovation. The important thing is being able to understand the technology and being able to use it for both benefit and risks assessments.”

Similarly, Hunter says that big pharma companies need to be quick to adapt before they are left behind by more innovative start-ups.
“I think pharma leaders need to think about what the world is going to look like in five to ten years’ time. If machines are reading everything what is it we want our scientists to do with that information? And then how are we going to use that information to be more efficient, innovative and successful?

“Bigger companies could buy AI start-ups but if they don’t have the internal systems to really maximise the adoption of what the start-ups have they’ll kill the golden goose. You need to have an environment where that AI start-up can continue to deliver and not meet resistance from internal structures.

“Of the pharma companies we’ve spoken to, some really get it, and some don’t. And within those companies in the middle it’s usually one particular division that gets it. But if we can show companies that this can be successful, I think there will be both a patient push and a payer push for AI in drug development.”

Even so, Hunter says it still won’t be long before AI in R&D is the norm rather than the exception.

“It won’t be more than 10 years, but different areas will adopt things at different speeds. It’s hard to predict, because I think a lot of it will be driven by where the early successes are. But is the technology here now? Absolutely.”