Science

Institut Pasteur and IBM use AI to accelerate drug discovery against Covid-19

The Institut Pasteur, specializing in the prevention and treatment of infectious diseases, and IBM have set up skills sponsorship to accelerate the search for therapeutic treatments in the context of the Covid-19 pandemic using artificial intelligence. The goal is to find an antiviral to prevent the virus from entering human cells. This requires finding drug candidate structures.

We wanted to help in the fight against Covid-19, says Xavier Vasquez, director of IBM’s technology centers. We knocked on the door of the Institut Pasteur to find out how we could help themThe Structural Bioinformatics Unit, headed by Institut Pasteur Technology Director Michael Nilges, responded favorably to this request.

Predict the binding potential of a drug
At that time, scientists from the unit had already initiated “the development of the architecture of prediction models of particular functional sites for therapeutic purposes“, explains Olivier Sperandio, researcher in the Structural Bioinformatics unit at the Institut Pasteur. With his team, he developed the” InDeep “system which is capable, from the 3-dimensional structure of a protein. -target, of “examine and analyze the folding, the three-dimensional shape and the presence of cavities“to predict”the binding potential of a future drug“.”The arrival of IBM has allowed us to scale up in terms of development and optimization“, adds the scientist.

It was the “Deep Search” tool developed by IBM that particularly interested researchers. This algorithm is able to perform specific queries in millions of scientific publications to extract precise information. “A search that would take 10 to 20 years could be reduced to a year, a few months or a few weeks“, details Xavier Vasquez.

Already widely used in the materials, automotive and energy industries, this tool has many advantages in the context of the Covid-19 pandemic, which has seen the number of scientific publications explode. The Express reported last June that since the start of the year, 20,000 studies related to Covid-19 had already been published.

Extract information and cross-check it
For example, it is possible to ask Deep Search which drugs have already been used so far and what the results have been. “It’s more than just a search engine. Deep Search is able to read tables, extract figures … to be able to link articles to each other and deduce a certain number of hypotheses.“, specifies Xavier Vasquez.

These capabilities won over the Structural Bioinformatics unit. “Deep Search can extract information about small molecules from documents, such as patents. Which is very complicated to do because they are embedded in tables“, says Olivier Sperandio.”We realized that the technology we were developing (InDeep, editor’s note) could be perfectly suited to this technology, that is to say to cross the world of data to which Deep Search could give us access with the world of prediction functional sites“, continues the researcher.

In practice, one of the objectives of the Institut Pasteur is to develop an antiviral which targets the proteins essential for the SARS-CoV-2 cycle. This type of treatment works at a specific point in the virus’s replication cycle to block it. In the case of Covid-19, it is the Spike protein (or S protein) that allows the virus to enter human cells via the ACE2 receptor, a protein found on the surface of cells in the respiratory mucous membranes. “It is through this transient interaction that the virus will invade the organism“, explains Olivier Sperandio.

Encouraging first results
The objective is therefore to find “a small molecule that would prevent the two proteins (Spike and ACE2, editor’s note) from talking to each other“. A particularly difficult, long and tedious task hence the use of artificial intelligence which makes it possible to automate part of the work. InDeep has thus made it possible”to identify functional sites on the surface of the ACE2 and Spike proteins. Then we use this information to better calibrate the identification of small molecules for therapeutic purposes.“. The objective is to identify areas essential to”the pathogenicity of the virus“on which would be fixed an antiviral treatment.

But there is still a long way to go before an antiviral can be marketed. Although the use of AI shortens some deadlines and the arrival of a vaccine against Covid-19 has been exceptionally rapid, “drug development takes a very long time. It takes about 15 years “, recalls Olivier Sperandio. “It’s still science fiction“, he adds, not wishing to comment on a future marketing date.

Back to top button