This startup used AI to design a drug in 21 days

[ad_1]

<div _ngcontent-c15 = "" innerhtml = "

microscope with laboratory glassware — Insilico Medicine aims to bring in-depth learning to the drug discovery process.

Getty

Insilico Medicine, based in Hong Kong, published a study Monday showing that its deep learning system could identify potential treatments for fibrosis. This system, called Tensor Generative Learning, or GENTRL, found six promising treatments in just 21 days, one of which yielded promising results in an experiment involving mice. The research has been published in Nature Biotechnology, and the code for the model was made available on Github.

"We have an artificial intelligence strategy combined with an artificial imagination," said Alex Zhavoronkov, CEO of Insilico, who compares GENTRL's operation to AlphaGo's machine-learning system developed by Google's Deepmind to challenge Go champion players.

Zhavoronkov founded the company in 2014. He previously had a background in computer science. He spent several years at ATI until his acquisition by AMD in 2006. He then changed course and decided to engage in research in biotechnology with a particular interest. in research on slowing down the aging process. He obtained his master's degree from Johns Hopkins, then a Ph.D. from Moscow State University, where he focused his studies on machine learning on the physics of molecular interactions in biological systems. He then worked for a number of companies, then returned to Baltimore to found Insilico. & Nbsp;

The original philosophy of the company was to use deep learning to form neural networks to scan large libraries of molecules to find drug targets. Shortly after the founding of the company, however, Zhavoronkov became fascinated by Ian GoodfellowWork in machine learning and decided to change course. & Nbsp;

"We thought," Can we make machines imagine new molecules with unique properties instead of filtering large libraries of suppliers? "In his opinion, the screening of molecules is what they do in the traditional world of drug discovery, but he wanted to know if this type of machine learning could do things faster. . & Nbsp;

Initial research published by the company around this idea in 2016 has generated investments in the competitive areas of biotechnology and AI. According to Pitchbook, it has so far collected $ 24.3 million worth of investments valued at $ 56 million, including from lenders, including A-Level Capital and Juvenescence. It also has many partners in the biotechnology field, including A2A Pharmaceuticals and TARA Biosystems. & Nbsp;

The current paper finds its origin in a challenge launched to society by its colleagues in the world of chemistry. They are asking the company to use its system to develop drugs that may inhibit the receptor activity of the domain of discoidin 1 (DDR1). DDR1 is an enzyme involved in fibrosis and, although it is not yet clear whether it regulates these processes, its inhibition is currently under study. This challenge was based on recently published research from a Genentech team, which took about 8 years to identify promising inhibitors of DDR1 kinase. & Nbsp;

An overview of the Insilico discovery process

Insilico Medicine

Insilico used GENTRL to design new drug candidates, which were then synthesized and a leading candidate was successfully tested in mice. It took about 21 days for the AI system to design molecules, and the total time required for design, synthesis, and validation was about 46 days. Although none of the GENTRL-designed drugs appears to be more effective than the inhibitors developed according to the traditional research method, the traditional process of developing drug candidates has taken over 8 years and millions of dollars to develop – compared to the handful of weeks and about $ 150,000 cost of the Insilico method.

"Their molecules are incredible, they are a little better than what our artificial intelligence could do," says Zhavoronkov. "But again, it's been years since people who do not have a lot of chemistry knowledge do that sort of thing."

Although he warns that Insilico still has a lot of work ahead of him, this research constitutes for Zhavoronkov an important advance, as it shows the chances of AI for drug discovery. & Nbsp;

"I think this paper will dispel a lot of skepticism in the big pharma companies," he says.

Insilico Medicine, based in Hong Kong, published a study Monday showing that its deep learning system could identify potential treatments for fibrosis. This system, called Tensor Generative Learning, or GENTRL, found six promising treatments in just 21 days, one of which yielded promising results in an experiment involving mice. The search was published in Nature Biotechnology, and the model code is available on Github.

Zhavoronkov founded the company in 2014. He previously had a background in computer science. He spent several years at ATI until his acquisition by AMD in 2006. He then changed course and decided to engage in research in biotechnology with a particular interest. in research on slowing down the aging process. He obtained his master's degree from Johns Hopkins, then a Ph.D. from Moscow State University, where he focused his studies on machine learning on the physics of molecular interactions in biological systems. He then worked for several companies, then returned to Baltimore to found Insilico.

The original philosophy of the company was to use deep learning to form neural networks to scan large libraries of molecules to find drug targets. Shortly after the founding of the company, however, Zhavoronkov became fascinated by Ian Goodfellow's work on machine learning and decided to switch courses.

"We thought," Can we make machines imagine new molecules with unique properties instead of filtering large libraries of suppliers? "He said that screening molecules is what they do in the traditional world of drug discovery, but he wanted to see if this type of machine learning could do things more." quickly.

The current paper finds its origin in a challenge launched to society by its colleagues in the world of chemistry. They are asking the company to use its system to develop drugs that may inhibit the receptor activity of the domain of discoidin 1 (DDR1). DDR1 is an enzyme involved in fibrosis and, although it is not yet clear whether it regulates these processes, its inhibition is currently under study. The challenge was based on recently published research from a Genentech team, which had taken about 8 years to identify promising inhibitors of DDR1 kinase.

Although he warns that Insilico still has a lot of work ahead of him, this research constitutes for Zhavoronkov an important advance, as it shows the chances of AI for drug discovery.

"I think this paper will dispel a lot of skepticism in the big pharma companies," he says.