Recognition of Quadruplexes by Methods of Deep Learning in the Mouse Genome

G-quadruplexes (G4) are nucleic acid sequences enriched with guanine and capable to form four-chain structures. Quadruplexes are actively studied, but the definite laws of their formation are still unknown, therefore, the problem of G-quadruplexes identification by deep learning methods plays a significant role in this phenomenon research. The solution of this problem by machine learning methods is complicated by the imbalance of the sample. The subject of this work is to generate quadruplexes with Generative Adversarial Network and add the generated data to a training set in order to improve the quality of various machine learning models for quadruplex recognition. This research demonstrates the results of four variations of Generative Adversarial Network (simple GAN, WGAN, WGAN-GP and LSGAN) on six combinations of convolutional networks from a generator and discriminator. CNN with a maximum ACCURACY of 0.96. The best and the most stable result of all the implemented architectures for this problem was shown by GEN2 + DISC1 with WGAN-GP. This research is one of the pioneering studies in the application area of Generative Adversarial Network networks in genomics problems and particulary in quadruplex recognition problems.