Get best of the week

Startup uses AI to search for molecules that will help fight coronavirus

Hello everyone. Right now, OTUS has opened a recruitment for the Python Neural Networks course , and today we want to share with you a translation of an interesting article about a company where our course teacher Arthur Kadurin works as chief AI officer .



Insilico Medicine, a startup based in Rockville, Maryland, claims that it used artificial intelligence to quickly identify molecules that could form the basis of an effective treatment for coronavirus infection at its peak.
It took 4 days for the Insilico AI system to identify thousands of new molecules that could be turned into potential anti-virus drugs. In Insilico say that they synthesize and test the 100 most promising candidates as well as publicly available library of new molecular structures, so that other researchers could also use them in their work.

UFO Care Minute


COVID-19 — , SARS-CoV-2 (2019-nCoV). — , /, .



, .

, , .

: |

The worldwide health emergency caused by the proliferation of the new deadly coronavirus known as 2019-nCoV has proven to be a real challenge for many modern biomedical technologies, new organizations and funding agencies that are striving to drastically reduce the time needed to create new vaccines and medicines to fight the pandemic.

The basic idea is to find and test new treatments by testing people for just one to two weeks instead of doing this for years. Gillead, a US biotechnology company, makes a dealwith a Beijing hospital earlier this week to start immediately testing the existing antiviral drug Remdesivir in Wuhan, the heart of the coronavirus outbreak, in humans.

Insilico, on the other hand, decided to check to see if they could find any clues that could help treat 2019-nCoV until January 28th.

Alexander Zhavoronkov, the founder and CEO, commented on it this way: “When a virus outbreak occurred, we did not realize how serious this is.”

According to Zhavoronkov, after the company decided to take part in the study, they studied a long list of all kinds of 2019-nCoV treatments that were published by the Beijing Global Health Drug Discovery Institute . As their goal, they chose an enzyme called a 3C-like protease, which is crucial for the reproduction of the virus.

Zhavoronkov said Insilico chose this target in part because it was similar to other viral proteases whose structures had already been modeled previously, and also because they had access to the model of the 3C-like protease 2019-nCoV developed by Rao Zihe , renowned expert on viral protein structures at Shanghai University of Technology.

Starting January 31st, Insilico has deployed 28 different machine learning models to develop new molecules that can bind to a 3C-like protease and inhibit its function.
Some of these methods use generative adversarial networks (or GANs), the same type of machine learning commonly known for creating deepfakes. But in this case, instead of generating highly realistic fake videos, AI generates new molecules that form a suitable structure for binding to the protease.

Insilico additionally uses machine learning methods to filter molecular structures that produce GANs: they give preference to structures that have “medicinal” properties and are chemically active, while at the same time discarding molecules that, judging by their properties, are unlikely to work as drugs, for example, metal compounds.

A number of filters are also used to help make sure that the set of generated molecules is not similar to existing known structures (Zhavoronkov claims that not a single molecule generated by his system matches the already existing one by more than 70%). The molecules are different from each other, so the company has a decent set of candidates for testing.

Four days later, Insilico software created hundreds of thousands of new molecular structures and filtered them, leaving several thousand that met the criteria that determine whether a potential candidate could become a drug.
“In four days we generated pretty good molecules,” comments Zhavoronkov .

The company published an article detailing its research in the free, non-peer-reviewed research repository Research Gate . The company also published its research and the structure of all potentially useful molecules on its website . Insilico encourages researchers to study and critically analyze the molecules created by his system, in the hope of speeding up the search for those molecules that could be useful for treating coronavirus.
Insilico is not the only company hoping that artificial intelligence will help offer new treatments for the Wuhan coronavirus. A team from the University of Michigan recently also published an articleabout using machine learning methods to create new drugs against 2019-nCoV.

Founded in 2014, Insilico has attracted about $ 50 million in venture capital financing today. It uses a number of different artificial intelligence-based technologies to develop new molecules that can form the basis of pharmaceuticals for treating diseases and predicting clinical trial results. The company partnered with pharmaceutical giant GSK and China's Jiangsu Chia Tai Fenghai Pharmaceutical to help them develop molecules for potential new drugs.


: «Triplet Loss: ».

More articles:


All Articles