SAN FRANCISCO—We certainly like to offer a full breakfast here at FierceBiotech, especially at the J.P Morgan Healthcare Conference, where attendees have probably had to take out a second mortgage for a three-day hotel stay.
So, we packed in as much as we could over a few hours early on Tuesday morning, the second day of JPM 2019, with a panel discussing what we see as the one of the biggest and potentially most complex issues for 2019: "Rewiring R&D: The Promises of Digital, AI, & Machine Learning for Biotech Research.
Our panelists: Andrew Allen, M.D., Ph.D., co-founder, president and CEO at Gritstone Oncology; Daphne Koller, CEO at Insitro; John Tsai, head of global drug development at Novartis; Alexis Borisy, partner at Third Rock Ventures; Kevin Julian, senior managing director of Life Sciences at Accenture; Saurabh Saha, global head of translational medicine at Bristol-Myers Squibb; and Tony Wood, SVP medicinal science and technology at GSK. They were all on fine form in front of a packed crowd, led by moderator and FierceBiotech editor Amirah Al Idrus.
For years, there’s been a lot of talk about artificial intelligence (AI), machine learning (ML) and digital in general around biopharma R&D, but often it can get trapped in a loop of hyperbole and big promises. We wanted to cut through some of the ambiguity around just what it means, how it can help drug development and ultimately, patients.
First up was Koller, the CEO and founder of Insitro, a company using machine learning for drug discovery and development. Koller herself used to work at Alphabet’s anti-aging biotech Calico, and was the Rajeev Motwani Professor of Computer Science at Stanford University, where she was hired in 1995 as the first machine learning professor and served on the faculty for 18 years. She was also named as one of Time's most influential people back in 2012.
She started off by trying to cut through some of the issues around these terms: “There’s a little bit of a confounding of terminology: There’s AI, and there’s machine learning, and some people think they are the same thing, but they’re not. There’s a large overlap between two technologies; you can think of artificial intelligence as the goal of trying to have machines do well what people do well.
“Machine learning, meanwhile, is the problem of building predictive models where you’re trying to pick a whole bunch of data and build a model that can predict something from a particular input. The problem of predicting of what’s going to be presented on a cell’s surface, for instance, is not AI: Because people can’t actually solve this problem.
“On the other hand, [Apple’s helper app] Siri is AI, but isn’t a machine learning-based application. So, they are two overlapping technologies, but in many ways, ML is the most useful way of solving a problem that seems to require intelligence.”
She said that using machine learning is like looking into a crystal ball to answer a question about a trial you might not be able to fully undertake. So, she said, as an example, if you want to predict what will be presented on a cell surface, but you don’t have enough tissue to actually measure that, you can turn to machine learning.
“If I have a sufficiently large dataset, which I can then train a model upon, then that ‘crystal ball’ is something I can use in clinical practice. And that’s a fundamental, core technology which can be used across many drug development settings, because there are so many scenarios where you wish you had this crystal ball of ‘what will this experiment do’ if only it wasn’t so expensive, or too long, and so on.”
Accenture's Julian said just listening to Koller revealed some of the issues biopharma faces, when terminology still needs to be explained and confusion still crops up. “If we ran the clock back five years ago, we wouldn’t be up talking about AI, we’d be talking about big data. It became this proxy where if you went back 10 years, no one was talking about big data. But then when it did come, every presentation you saw from every company would have this term plastered across its slides: We have big data, we’re processing big data. Everyone wanted to have big data, but nobody knew what it was.
“AI today seems to have taken that same role; so now, every presentation will have AI on it. And one of my favorites is AI/ML. That basically says: I don’t know what it is, but I know I need to have it, and I’m just going to put it all there, and I can say: We got it! But you see company pitches with these terms in and you’re like: What are you talking about? So, there is a communication problem here,” Julian said.
“I would encourage all of us in the field to realize these are all very powerful tools. Daphne just said it very well: There are lots of very real examples where you can have predictive models that are highly effective, and there are enormous efficiencies to built into what we do.”
Third Rock's Borisy said that while we can get bogged down in the terminology, these are tangible; they are real. He explained: “There are crazy expectations from these, but there has been rapid increase in investment, something like 60% CAGR investment in the last five years in these technologies, so there’s something here. The challenge for us is to get over that peak and through the reaction of the industry to it, and we’re starting to see of the reaction to the attempts in the marketplace to hype AI.
“But this isn’t that much different to big data or the cloud or even back to EDC: People got hyped up by it, got really excited by its promise, but maybe we got a little ahead of ourselves, but we are getting through to the point where the productivity is real.”
GSK's Wood said the important thing is to start with what problem you are trying to solve. “From that, you then need to determine whether you can generate the appropriate dataset for that problem. And then from that, what method am I then going to use to interrogate that data.”