Time For AI To Deliver In Drug Discovery, Says Atomwise CEO
Executive Summary
Having signed a multi-year, multi-target agreement with Lilly worth potentially $550m, Atomwise CEO Abe Heifets tells Scrip the burden of proof is on AI practitioners to start producing relevant and actionable data.
Eli Lilly & Co. is the latest big pharma to unveil a major artificial intelligence (AI) drug discovery collaboration after signing an agreement with one of the pioneers in the field, San Francisco-based Atomwise Inc..
The companies will collaborate on up to 10 drug targets selected by Lilly, with the goal of accelerating the time it takes to identify and develop potential new medicines. Atomwise could receive up to $1m per target in discovery milestones and will be eligible for up to $550m in development and commercialization payments.
In an interview with Scrip, Atomwise CEO Abe Heifets acknowledged that "there's a ton of excitement about AI which is a very broad space in the same way that pharma is broad. We're seeing many different applications, everything from patient diagnosis to basic biology, but we focus on the chemistry side. There are applications throughout the process which is part of the reason why you see such interest."
There are a lot of AI companies targeting the pharma industry, which is looking for ways to lift its traditionally low success rates for bringing drugs from discovery through clinical development and onto the market. The hope is that new technologies can improve those success rates, but Heifets noted that it is not easy.
"Nobody can take a look at an algorithm and say, yes of course, for this particular problem, this particular algorithm will be the best. I don't think anybody can just do it from the underlying math," he said. "Like so much drug discovery, it's an empirical science and you have to just go and demonstrate success after success after success. You have to show where it works, where it doesn't, you have to convince people through data, and I think the burden is on us, the AI practitioner, to provide that evidence – and that's what we've done."
Atomwise said that it can analyze a very large chemical space involving billions and billions of compounds to identify a small subset with high specificity for synthesis and testing. The company, which was founded in 2012 and invented the first deep learning AI technology for structure-based small molecule drug discovery, added that processes that traditionally take years can be compressed with Atomwise’s technology to a matter of weeks.
Heifets said that the projects the company is already running showed that "our approach is working." He cited an alliance with the Drugs for Neglected Diseases initiative (DNDi), whose scientists selected three "verified but challenging" therapeutic protein targets that would inhibit the action of the parasite that causes Chagas disease. For each disease protein, Atomwise screened millions of compounds to predict those that bind and potentially inhibit protein function. In April the partners announced that the research had delivered drug-like compounds that would go on to further optimization and then potential development.
A partnership with the University of Connecticut has seen the discovery of inhibitors of the protein that becomes over-activated during ischemic strokes, Heifets noted, pointing out that as well as the Lilly pact, which is focused on developing drugs for novel target proteins that "are often challenging and less well studied,” Atomwise is also working alongside other big pharma players, notably Merck & Co. Inc., AbbVie Inc. and Bayer AG, "and we're showing that in a wide range of disease areas that it works."
The company also linked up with the contract research organization Charles River Laboratories International Inc. in January with the aim of making "historically intractable targets become new therapeutic opportunities." The total potential value of the royalties to Atomwise with success in all projects could exceed $2.4bn.
The Lilly deal, which gives Atomwise the option to develop compounds that Lilly chooses not to advance into clinical testing, is a welcome boost financially for a company which has raised $51m so far, $45m of which came in a series A financing in March 2018. Heifets said that the new partnership "really underscores our belief that AI is the industry standard but as I said, it's not easy and the true mark of success is going to be the eventual development of new therapies and new options."
In a recent interview with Scrip following the signing of a deal with the UK's BenevolentAI, Mene Pangalos, head of R&D biopharmaceuticals at AstraZeneca PLC, noted that "there is a lot of hype about machine learning" and while he is optimistic about how AI could indeed transform drug discovery and development, "right now it's a hypothesis. I'm not going to say it will be very useful until I have got some evidence that it has actually generated something that we wouldn't have otherwise done on our own." (Also see "AZ Inks Machine Learning Deal With BenevolentAI" - Scrip, 30 Apr, 2019.)
Heifets thinks that is a fair point, saying this is "both a nice and challenging thing and we've got to raise the table stakes, go out for a couple of dozen projects where nobody knew what the right answer was, find the right answer and show it was the right answer. In AI, there is a lot of predicting yesterday's stock price but it is much more interesting if you can predict tomorrow's."
He concluded by saying that "scientists are willing to be persuaded by data, so you have to show them data and then they're willing to believe." Atomwise has run over 200 projects and "we have had successes in over 60% of them, a significantly higher rate than with other technologies."