Nikolay Kutuzov, a member of the laboratory team, won the hackathon «Digital Pharmacology: Predictive Modeling», which took place from April 28 to 30. 

Participants based on open sources using artificial intelligence technologies competed in building a machine learning model to predict the parameters characterizing the toxicity of a chemical compound. The prize fund of the event was 1,000,000 rubles. The team consisting of Nikolai Kutuzov and Sergey Novikov took the first place.

Kutuzov and Novikov's solution is based on a unique author's approach to solving the problems of analyzing chemical compounds — aggregation of features from various sources and application of the gradient boosting algorithm of trees to them. These are both chemical signs (the greatest increase in quality among them was given by counting functional groups in the molecule), and data from language models. Data from 20 sources made it possible to form a dataset of more than 1.1 million molecules to predict 34 signs. Having received a very complete feature space, the gradient boosting algorithm segments it into blocks, in each of which an average prediction is output.

Source: news note about the results of the first joint event of the ITMO x Synthelli x Medtech Infochemistry Center.Moscow.