In silico моделі прогнозування мутагенності Еймса факторів навколишнього середовища

Background. The classical in vitro and in vivo methods developed and widely used in the past decades to assess the genetic effects of environmental factors are complex in view of their implementation, are expensive, long-lasting, have the problem of reproducibility of the results of experiment in different laboratories and may face ethical problems of using warm-blooded animals in experiments. Objective. Development, optimisation and testing of effective in silico models for assessment of Ames muta­genicity of environmental factors. Methods. The genetic assessment of the impact of environmental factors was carried out in accordance with a set of chemical compounds for which information on potential mutagenic activity was obtained experimentally, using the in vitro Ames Salmonella/microsome test. Four machine learning models were developed to solve the problem of binary classification to form two classes of xenobiotics (mutagen/non-mutagen). The total sample is represented by a set of 8,083 xenobiotics. Results. We developed four machine learning models with 85% accuracy, matching the reproducibility of Ames test data across laboratories. In addition, we have proposed a binary classifier that subject to dimensionality reduction of the input data, taking into account the qualitative composition of molecular descriptors, allows us to improve the accuracy of in silico prediction of genotoxicity of chemicals. Conclusions. The necessity of updating and expanding the list of effective and more productive methods and approaches for assessing the genotoxic effects of environmental factors is substantiated, which allows avoi­ding the use of warm-blooded animals in the experiment, saving time and reducing the number of false-negative and false-positive results. The possibility of increase the accuracy of predictive machine learning models for assessing the genotoxic potential of environmental factors in conditions of dimensionality reduction of the data set is presented.