Aller au contenu principal

Using AI to identify anti-viral potential of Covid-19 treatments

Research Article published on 15 December 2020 , Updated on 22 January 2021

With around 3,000 people still hospitalised in intensive care units and more than 58,000 deaths in France due to Covid-19, the search for an effective treatment for the disease remains of paramount importance. Artificial intelligence is being used by researchers at Université Paris-Saclay to guide their research.

 

It is often necessary to look back, learn and be inspired by what already exists in order to develop new antiviral treatments. As clinical trials are usually long and at times inconclusive, it is natural to look to drugs that are already known and to try to reposition them to treat Covid-19. However, the databases of therapeutic molecules are vast, so how can the right treatment be found quickly? This is what Émilie Chouzenoux, from the OPIS joint-team (Inria, Centrale-Supélec) and her colleagues from Indraprastha Institute of Information Technology (IIIT) in Delhi and the Institute of Post Graduate Medical Education and Research (IPGMER) in Kolkata have been pursuing in their recent work.

 

AI provides advice

The team has developed a tool capable of speeding up the race for antiviral treatments. Rather than launching a multitude of clinical trials in search of an effective treatment and manually reviewing the available drugs to check their effectiveness, their artificial intelligence (AI) tool “learns” from what is already known to identify the antiviral potential of molecules. The principle governing these calculations is very similar to that of web-based referral systems.

The model cross-references databases of genetic sequences of known viruses and chemical structures of drugs already in use. The similarities existing between virus/drug pairs and measured thanks to the frequency of oligonucleotides (OligoNucleotide Frequency, ONF) for viruses and the SIMCOMP index (SIMilar COMPound) for molecules, can then help to find a treatment which is likely to be effective.

This model has proved to be successful when applied to SARS-CoV-2. Of the six molecules selected by the tool, four have been or are still in clinical trials, namely Remdesivir, Ribavirin, Sofosbuvir and Umifenovir. 

 

Antiviral research in the future

This technology paves the way for targeted clinical trials and could improve their success rate. The reaction to a new virus or a new mutation of a virus would become faster and more effective. With a much higher mutation rate than their DNA counterparts, RNA viruses such as SARS-CoV-2 often make the search for treatment harder. However, by adjusting its response to each new strain of the virus, AI can keep up.

The work by Émilie Chouzenoux and her colleagues, although certainly motivated by the Covid-19 pandemic, seems to be an encouraging solution on a larger scale. It is with this in mind that the team has shared its database and codes. At a time when viruses threaten to become more numerous and more resistant, biomedical research now has an additional and promising tool at its disposal. 

 

Publications :