Prediction of 3D Сhromatin Structure with Machine Learning Methods

In recent years, applications of Machine Learning have substantially increased. In particular, the utilization of these methods in the field of molecular biology is especially significant. The development of technologies enables rapid generation of large amounts of epigenetic data. The novel Hi-C technology provides an opportunity to obtain data on the genome interactions that has unraveled many principles of chromosomal folding, including the subdivision of the genome into Topologically Associating Domains (TADs). Several studies have confirmed the correlation between the structure of chromatin and other epigenetic features [4, 5, 7]. My research focuses on using Machine Learning methods to predict the three-dimensional structure of chromatin using ChIP-seq epigenetic data on chromatin markers. This paper considers linear ​models with three types of regularization and an architecture of Recurrent Neural Networks. As a result, predictive models were trained and evaluated, the best score of weighted MSE was performed by Neural Networks. Informative epigenetic features were identified that leads to studying their biological significance.