AI study led by Insilico's Alex Zhavoronkov bolsters case for faster, cheaper drug discovery
Drug development is an arduous and expensive business. The promise of artificial intelligence is that machines can wean manufacturers away from the breadth of resources it takes to discover a potentially potent compound, and leverage the medical conditions it could be used to treat. Alex Zhavoronkov, whose Hong Kong-based AI-shop Insilico Medicine has entrenched itself in the tentacles of biopharma R&D, now has some data to bolster the fervor and investment that AI has drummed up.
On Monday, Zhavoronkov and his fellow AI scouts from Wuxi and scientists from the University of Toronto, published a paper in Nature Biotechnology that supported the feasibility of employing a machine learning approach — generative adversarial networks (GANs) and reinforcement learning — for de novo drug design. The researchers found that they were able to identify a myriad of compounds targeting a protein called Discoidin domain receptor 1 (DDR1) — which is expressed in epithelial cells and involved in fibrosis — in a swift 21 days.
After that, six molecules were then selected for synthesis in the lab — those tests revealed four with potential. Further testing whittled it down to one compound, which was tested in mice. That data suggested the molecule conferred a potent effect against the protein — although much like any other compound, its safety and efficacy must be validated in human trials.
This study, which was conducted in response to a challenge set by a partner company, is a taste of things to come, Zhavoronkov noted in an interview with Endpoints News. The plan was to take an algorithm developed a few years ago — an algorithm that is openly available, and use it as a proof-of-concept to demonstrate the potential for AI in drug discovery, he asserted.
“Imagine…Pfizer putting out all of their preclinical stuff for free for everybody to use, right? They won’t do it. Or J&J putting out all of their data, even the old stuff? They just don’t do it. We decided to kind of do a demo. So the skeptics are… a little bit less skeptical.”
There's been a lot of buzz about #AI for drug discovery. But I think this is the best case yet: All in 46 days!https://t.co/BnfK8VcGek@InSilicoMeds @uoft @VectorInst @biogerontology @A_Aspuru_Guzik and colleagues@NatureBiotech #DeepLearning pic.twitter.com/eHtOhaaksA
— Eric Topol (@EricTopol) September 2, 2019
Recent estimates by the Tufts Center for the Study of Drug development maintain that taking a drug all the way from discovery to approval costs roughly $2.6 billion (in 2013 dollars). Steven Paul, current Karuna chief, published a study in 2010 in which he highlighted the magnitude of resources it cost his former employer Eli Lilly $LLY to discover new compounds: the out-of-pocket cost for lead optimization came to a hefty $146 million.
“In this work, we designed, synthesized, and experimentally validated molecules targeting DDR1 kinase in less than 2 months and for a fraction of the cost associated with a traditional drug discovery approach. This illustrates the utility of our deep generative model for the successful, rapid design of compounds that are synthetically feasible, active against a target of interest, and potentially innovative with respect to existing intellectual properties,” Zhavoronkov et al wrote in their paper.
Zhavoronkov compared the study to AlphaGo — the first computer program developed by Alphabet’s AI company DeepMind and immortalized in a Netflix documentary — to defeat a professional human Go player. Go, which originated in China over 3,000 years ago, is a deceptively simple strategic thinking board game that has an incredible 10 to the power of 170 possible board configurations (which is more than the number of atoms in the known universe.)
“We kind of thought about this paper as a kind of mini AlphaGo,” Zhavoronkov said. “I hope that the big executives will also kind of hear this and understand that…we actually put the entire discovery process on display.”
“Every second student in China wants to be an AI scientist…this (AlphaGo) miracle might not have impacted their lives significantly, but it really changed the mentality for everybody,” he added. “So that is what we need in pharma. People should focus less on geopolitics or…warfare. This is the stuff to watch right on TV. This is the cool thing.”
A platoon of biopharmas have linked up with the emerging crop of AI specialists itching to capitalize on how large datasets can be harnessed to drive new therapies into the clinic. Zhavoronkov is well connected — last year he raised funds at the behest of Shanghai high-flyer WuXi AppTec, Singapore’s Temasek, Peter Diamandis and Juvenescence.
Isolating compounds for development is one aspect of the ballooning AI industrial complex — it also has appeal in another issue drug developers regularly contend with — the low odds of success, even with compounds that show great potential in early testing. Then there’s the golden question — even if AI can help make the process of drug development better, faster and cheaper — will that translate to less expensive treatments?
Handy hint to anyone pitching me a story saying that AI is cutting the time needed, and cost of. developing drugs: tell me how this will cut drug prices.
— Natasha Loder (@natashaloder) September 2, 2019