Artificial Intelligence, BioPharma

Watson for Drug Discovery may be down, but AI in biopharma isn’t out

Last month, it was reported that IBM would stop selling Watson for Drug Discovery, but executives in the field of biopharma artificial intelligence say the field is still going strong.

AI, machine learning

The news last month that IBM would stop selling Watson for Drug Discovery due to lackluster financial returns did not come as a surprise to some executives in the field of artificial intelligence-led drug discovery and development. But, they said, that doesn’t mean the field as a whole is in trouble.

STAT reported April 18 that IBM would continue servicing existing customers, but would otherwise stop development and sales of the product, which the tech giant launched under a partnership with drugmaker Pfizer in 2016.

The launch of IBM’s Watson for Drug Discovery occurred as the idea of applying artificial intelligence and machine learning to the biopharma industry – particularly to aid discovery of new compounds – was beginning to take off in earnest. Since then, numerous companies have launched efforts of their own.

For one such company, Salt Lake City-based Recursion Pharmaceuticals, the news did not come as a major surprise. “Our general take has been that taking public data sets and literature as your base input is going to be incredibly difficult because of how non-relatable that data is and hard it is to reproduce,” Recursion CEO Christopher Gibson said in a phone interview.

A key distinction is between analysis of data sets via AI that is prospective versus retrospective. For AI to work, Gibson said, data sets need to be tailored to it. Like a post-hoc analysis of clinical trial data, retrospective analysis using AI of data sets that were not designed for it is not statistically satisfying. With that in mind, the company last week released open-source data from what it called the largest ever data set of biological images – totaling two petabytes, or 2 million gigabytes – in order to encourage companies to develop new machine learning algorithms in experimental biology and drug discovery.

“I think the first point, and maybe the critical one, is that drug discovery is hard,” said Abe Heifets, CEO of San Francisco-based Atomwise, in a phone interview. Atomwise has a deep learning-based AI technology it developed for structure-based discovery of small molecule drugs. The company partners with numerous other firms, ranging from large pharmaceutical companies like Merck & Co. to the contract research organization Charles River Laboratories and Tulane University. Earlier this month, Atomwise announced a partnership with Y Combinator, where the company itself was incubated, to make the drug discovery service it created available for YC companies for free.

Heifets said the success rates for bringing drugs from discovery through clinical development and beyond via traditional methods are generally pretty low. Hence, companies like his continue looking for new technologies to improve those success rates. “That’s the context through which we have to interpret this news” about IBM Watson, he said.

Photo: ANDRZEJ WOJCICKI, Getty Images

CLARIFICATION: When referring to success rates in drug development, Atomwise’s Abe Heifets was referring to traditional methods. The story has been updated.

Shares1
Shares1