Identifying Super-enhancers Using DNA Sequence Embedding Representations

Super-enhancer(SE)is a cluster of active typical enhancers(TE)with high levels of the master transcriptional factors,cofactors and histone modification marks.Super enhancers drive the expression of genes controlling cell identity and can be used to explain cell type-specific expression patterns.Currently scientists are mainly using various high-throughput data of different transcription factors or chromatin markers to distinguish super enhancers from typical enhancers.This experimental approach is usually costly and time-consuming.Therefore,it is necessary to build a computational model to identify super enhancers.In this paper,we propose a model named DeepSE which is based on a deep convolutional neural network model to distinguish super enhancers from typical enhancers within multiple cells in the mouse and human genome.DeepSE represent the DNA sequences using the dna2vec feature embeddings.Experimental results show that DeepSE outperforms all the existing state-of-the-art models when use DNA sequence information only,Besides,DeepSE can be well generalized across different cell lines.Although super enhancer is kind of cell-type specific element,but there may exist common and hidden sequence patterns that are shared across cell-type specific super enhancers.Using it as an alternative to traditional experimental methods can substantially reduce experimental costs.The source code and data are stored in the GitHub repo(https://github.com/Qiaoying Ji/DeepSE).