Recognition of Genome Functional Elements with ChiP-seq Experiment Distributions by Deep Learning Methods

This work is devoted to the search for patterns of dependencies between histone labels and secondary structures of DNA, which are quadruplexes and Z-DNA. Due to deep learning methods, more specifically convolutional neural networks, it becomes possible to train models that give a good quality of genome functional elements recognition. In this paper we, for the first time, will be using the data processing method to recognize patterns of histone marks and secondary DNA. This method was used in the DeepChrome developing, which improves the quality of features for training and avoids discretizing that had been used in earlier research. We were able to obtain results that overcome ones from the earlier studies in the same area. The maps of the signal strength of histone marks learned by the models allow us to identify the most important signals.