GAN Implementation to Patterns Exploring in the Secondary Structures of DNA

Annotation This work explored the possibility of using deep learning models to generate nucleotide sequences with certain biological characteristics. Throughout the study, both the ability to directly create DNA chains and predict their biological properties based on existing data were independently explored. The main goal of the work was to create an algorithm for expanding the data sample in a generative way without losing the biological meaning and to evaluate the prospects of this approach. The deep learning models used in this work are based on a residual learning architecture, trained using several techniques and modifications. During the work, the strengths and weaknesses of using a generative approach were discovered, and the results of the work opened up further opportunities for research. The ResNet-based architecture chosen for study showed itself well as a generative tool. A model capable of generating nucleotide segments with known features was developed, and a sampling expansion method was implemented, which made it possible to improve the quality of prediction of the existing model.