An artificial intelligence system has been created that calculates the results of organic chemical reactions

IBM researchers have developed a new artificial intelligence system based on the methods used to translate text from one language to another. However, this program treats the atoms of chemicals as letters, and molecules as separate words, and the result of its work is by no means a translation, but the definition of the final results of complex organic chemical reactions, which, in turn, can be used to speed up development new drugs, for example.

Over the past 50 years, scientists have tried their best to teach computers all the intricacies of the “work” of chemical transformations in order to obtain reliable predictions of the results of organic chemical reactions. However, organic compounds can be so complex that modeling their chemical properties and behavior during reactions requires an unacceptably high cost of computing resources, otherwise the results obtained have low reliability in order to be 100% reliable.

So researchers from IBM Research in Zurich decided to make a very unusual move. Using an artificial intelligence-based translation program, they adapted it to work in the field of organic chemistry. As in most other cases, the new program is based on an artificial neural network that has been pre-trained on millions of known chemical reactions, some of which are very complex and are carried out in 30-40 consecutive steps. In other words, the IBM scientists taught their translator software the basic structure of the “language” of organic chemistry.

Currently, the program provides the issuance of results, the reliability of which is at least 80 percent. At the same time, if artificial intelligence finds that a chemical reaction can go in different ways, each of which will lead to a separate result, it will give all the options for possible solutions, ordering them according to the calculated probability.

Until recently, the new artificial intelligence system had to deal with organic molecules, the number of atoms in which did not exceed 150. “However, there is no theoretical limitation that prevents us from working with longer and more complex molecules,” says Theophile Gaudin, one of the researchers.

In the near future, researchers at IBM Research plan to make their system available to everyone through a “cloud” service. By that time, according to available forecasts, the system will already ensure the reliability of the results of at least 90 percent. In addition, in the new version of the system, a number of additional factors will already be taken into account, such as the current temperature, pH-factor, used solvents, etc. At the same time, artificial intelligence will have to figure out most of the influence of these factors on the results of a chemical reaction on its own in the course of self-learning.

“In addition, we plan to hold a kind of competition in which our program will compete with groups of experts in the field of organic chemistry,” says Theophilus Godin, “However, this does not mean that artificial intelligence will replace chemists in the future, we originally created it as a tool that would facilitate and speed up their work. “

Source: www.dailytechinfo.org