AI and Big Data terms change their content all the time — this is an example of their ontological uncertainty. I was very much engaged in improving and inventing new methods of data processing in biomedicine and related fields (biochemistry, biophysics) — a large part of my PhD thesis is devoted to these topics. Then there was the previous wave of interest in neural networks — and in Pushchino, where I did my PhD, there were several conferences of young scientists devoted to the use of neural networks in various biochemical and biophysical applications.

At that time, neural networks were not so-called 'AI': it was a broader concept, it was 'machine learning'. Better results were obtained with other methods, but after that, I came to grips with molecular modeling — even then it was already tens, even hundreds of terabytes. Many issues resolved then are now the basis of methods for Big Data processing: modeling of multi-particle molecular systems, plasma; modeling of the development of the universe, where gravitational, Coulomb, potentials were taken into account. Especially after my work as the Director of the Supercomputer Centre in Glasgow, it became clear to me that these methods can be used in the most diverse areas.

And then the term "Big Data" came along. I blend with this tide, because on the one hand, I had decade-long mathematical experience, and, on the other hand, I was well-versed in its applications. Now it is more of a terminological rethinking, although I have not positioned myself as a data scientist. Now I'm comfortable with leading programs because I see applications of this foundation.

We aim to advance computational and machine learning techniques to a preclinical level of results readiness. Now the clinic sets limits, and we are trying to optimize the preclinical stage of testing and developing new drugs. There are certain successes — we can predict the toxicity and, what is most important, the transferability of results from animal models to humans. This is a real breakthrough: the results obtained for animals are not always applicable for humans, but it turned out that there are general principles. Plus, it is possible with a high degree of certainty to predict the toxicity of
a human drug using animal models.

This reduces the risks of clinical trials: volunteers are at risk, because animals' results do not always correlate with the humans. We reduce not only these risks, but also
the cost of the medicine by understanding what can be obtained from at an early stage of its development. Prospective toxic drugs will not even be synthesized.

The number of investments in this area has increased by 800% over the past year. Partly due to the pandemic, these methods have helped in the development of vaccines in a number of cases. Again, this is about data analysis, and this topic has been developing exponentially for a long time. More and more startups are appearing, for example, our startup "Syntelly" .

The development of drugs is a very expensive process, and reducing the timeframe even by a month is a worthy period. Anything that can reduce risks has great demand. Another question is the study of chemical space. The combination of mathematical modeling methods with machine learning helps to find fundamentally new structures, to find new structures with potentially low toxicity and high efficiency. We all know how inconvenient it is to handle vaccines and other molecules that are unstable at high temperatures. The chemical universe is huge — 10⁶⁰ compounds is an astronomical number.