CAMBRIDGE, Mass. — Biotech is famously skeptical of Silicon Valley pitches to fix all that ails drug discovery with a prescription for pairing big data and artificial intelligence.
But one of the pharma industry's top R&D chiefs anticipates the day is coming fast when advanced computing techniques like machine learning yield results for the sector.
"I would predict that in the next two to three years we'll have examples of targets, maybe not reagents, but targets discovered through machine learning and ‘in silico' methods that unravel biology that wasn't appreciated before," said Hal Barron, GlaxoSmithKline's head of R&D, speaking Thursday on a panel at the STAT Summit.
Drug discovery is time consuming, hampered by human biology and incomplete knowledge of disease states. The development process that follows it is highly capital intensive, making missteps early on exceedingly costly.
"It's a long journey where you have multiple forks in the road, multiple stages of drug discovery," said Daphne Koller, CEO of drug discovery startup Insitro and an expert in AI, at the Summit. "Ninety-nine percent of the paths are going to get you to a dead end and make it longer."
"If you had a somewhat accurate compass," she added, "think about what that would do to the probability of success of the process."
In general terms, that's the promise of using machine learning or other artificial intelligence approaches to comb through genomic databases, or draw links between molecular structures and drug activity.
In practice, AI has yet to make much of an impact on drug discovery success rates or speed, which makes Barron's prediction notable. And discovering new targets is only the start, of course. Examples abound of known biological targets that have been considered "undruggable" due to challenges in medicinal chemistry.
Under Barron, who joined GSK in January 2018, the British drugmaker has made a concerted effort to revamp its R&D efforts, trimming programs, staff and therapeutic focus. Barron also championed a deal with 23andMe that gives GSK access to the genetics company's large database — one which the pharma hopes to probe using machine learning.
"I don't think you put your toe in the water in this field," said Barron. "You have to believe this is a very transformational opportunity to completely rethink how drug targets are discovered."
While Barron appears to think that way, his philosophy may not be as widely shared across the pharma industry as the many breathless press releases might suggest.
"I think there are one or two other [pharmas] besides GSK where the people in leadership positions believe it's going to transformative," said Insitro's Koller in a separate interview with BioPharma Dive.
"Having enough conviction that you are going to rip things out by the root and do them differently is really, really challenging in an organization of 100,000 people."
More common than R&D overhauls are targeted collaborations between pharma and AI specialists, like the one signed between Insitro and Gilead to investigate new targets for NASH, a common liver disease.
A half dozen pharma companies, meanwhile, have appointed chief digital officers to their executive committees within the past two years in hopes of improving how they generate and use data, all with techniques like AI in mind.
But realizing AI's potential in drug discovery and, later, development, might take a new type of pharma company, Koller suggested in an interview.
"When I look at the big tech companies that have been successful, Amazon did not emerge from WalMart, Google did not emerge from Yellow Pages, Netflix did not emerge from Blockbuster," said Koller.
"So is it the case that a [company like] GSK could enact this kind of cultural transformation? Maybe, but you also want to have some newcomers."