The system will start working next year
Russian scientists from the Moscow Aviation Institute have developed an artificial intelligence model that will help check drugs for compatibility.
Photo: unsplash.com
It is not uncommon for patients to take several different medications. However, they do not always go well with each other. According to medical research, taking up to 5 medications at the same time entails a risk to health from their incompatibility in about 5% of cases, with 6 or more, this probability increases to 25%. Thus, in every fourth case, this leads to some negative consequences for health.
Until now, the problem of drug compatibility has been solved by a doctor accessing data from a special medical database — the «State Register of Medicines». But studying numerous medical instructions takes a lot of time.
With the advent of artificial intelligence technology, developments around the world have begun on special programs that would help doctors quickly resolve the issue of drug compatibility before prescribing them to patients.
For example, in the West, they are trying to teach AI to analyze the interactions between drug molecules.
According to Vladimir Sudakov, a leading researcher at the Keldysh Institute of Applied Mathematics of the Russian Academy of Sciences and professor at the Department of Computational Mathematics and Programming at MAI, he and his colleagues took a different path. Their approach to training an artificial intelligence model is based not on the analysis of chemical processes, but on the analysis of texts and large language models.
With this approach, artificial intelligence, without knowing the intricacies of chemical processes, very well studies the entire corpus of medical texts available to modern pharmacology, all kinds of instructions from different databases. Having “absorbed” hundreds of gigabytes of data, the supercomputer and a special program “remember” the entire set of data and draw correct logical conclusions based on them.
Test operation of the system showed that the model accurately indicates the correct result, but the most interesting thing is that it was able to establish a connection between drugs even where medical specialists had not previously seen it.
The system is currently undergoing test operation, and in 2025 it is planned to be implemented in the medical information system of one of the Russian clinics.