Year of Graduation
Methods of Machine Learning in Genetics and Genomics
Applied Mathematics and Information Science
In this paper we consider the problem of recognition of histone marks from the nucleotide sequence of DNA. The aim of the study is to identify new patterns of gene expression and their relation to the epigenetic code. The classification problem is solved by constructing a deep convolutional neural network. Six different histone modifications of H3K27ac, H3K4me1, H3K9me3, H3K4me3, H3K27me3, H3K36me3 are taken for research. Six models are considered, each trained on the peaks of its own modification taken from stem cells. ROC and Precision-Recall curves are constructed, AUC-ROC for all 6 classifiers is calculated. For comparison, the AUC-ROC of a random forest model is calculated on the same data. The effectiveness of the obtained models was also tested on cells of other tissues: nerve, bone, liver, lung and kidney. Neural network power indicators in the classification of stem cell sites on which the models were trained can provide new information about the degree of dependence of gene expression from epigenetic code. Checking AUC-ROC on other tissues will allow to determine whether there are some common patterns of expression regulation in different types of cells.