DeepHistone: a deep learning approach to predicting histone modifications

Abstract

Motivation: Quantitative detection of histone modifications has emerged in the recent years in a major means for understanding such biological processes as chromosome packaging, transcriptional activation, and DNA damage. However, high-throughput experimental techniques such as ChIP-seq are usually expensive and time-consuming, prohibiting the establishment of a histone modification landscape for hundreds of cell types across dozens of histone markers. These disadvantages have been appealing for computational methods to complement experimental approaches towards large-scale analysis of histone modifications. Results: We proposed a deep learning framework to integrate sequence information and chromatin accessibility data for the accurate prediction of modification sites specific to different histone markers. Our method, named DeepHistone, outperformed several baseline methods in a series of comprehensive validation experiments, not only within an epigenome but also across epigenomes. Besides, sequence signatures automatically extracted by our method was consistent with known transcription factor binding sites, thereby giving insights into regulatory signatures of histone modifications. As an application, our method was shown to be able to distinguish functional single nucleotide polymorphisms from their nearby genetic variants, thereby having the potential to be used for exploring functional implications of putative disease-associated genetic variants. Conclusions: DeepHistone demonstrated the possibility of using a deep learning framework to integrate DNA sequence and experimental data for predicting epigenomic signals. With the state-of-the-art performance, DeepHistone was expected to shed light on a variety of epigenomic studies.

Dynamics of gene expression and chromatin marking during cell state transition

“…The epigenetic mechanisms leading to the development of an individual or to the differentiation of a cell lineage from the unique genotype of the organism have been largely studied during decades. Although initial references to the mechanisms by which epigenetics promotes cell memory and leads cell fate did not relate to its ability to regulate gene expression, a causative role for epigenetic modifications in controlling transcription has been later pointed out (see Kouzarides, 2011, andRivera andRen, 2013, for reviews about different aspects related to epigenetics and its role in regulating gene expression), and it has even been shown that some epigenetic features, such as histone modifications, are accurate predictors of gene expression (Karlić et al, 2010;Dong et al, 2012;Sekhon et al, 2018) and the other way around (Yin et al, 2019).…”

“…In particular, a large fraction of genes in the human genome (likely more than 50%, Pervouchine et al, 2015) are either invariably silent and not marked, or expressed and marked across most cellular states. Genes with stable epigenomes and transcriptomes drive the correlations to large values when computed in a particular cell condition, and explain why models relating gene expression to histone modifications inferred in a particular cell type have high predictive power in other cell types (Karlić et al, 2010;Dong et al, 2012;Sekhon et al, 2018;Yin et al, 2019), even though there is no true causality involved in the relationship between chromatin and expression. The steady-state correlations represent an example of the Sympson's paradox (Simpson, 1951), by which the data can show different or even opposite behavior if subgroups within the dataset are considered.…”

Identifying complex motifs in massive omics data with a variable-convolutional layer in deep neural network

“…5-8 for more details), especially on datasets containing multiple combinatorially regulatory cis-elements of various lengths [38,48]. Meanwhile, vConv could be readily integrated into multi-layer neural networks, as an "in-place replacement" of canonical convolutional layer with potential applications in more scenarios such as cis-regulatory motif prediction [49], predicting non-coding functions de novo [50], predicting the transcription factors binding intensities [51], and quantitative detection of histone modifications [52]. To facilitate its application in various fields, we have implemented vConv as a new type of convolutional layer in Keras (https://github.com/gao-lab/vConv).…”

Deep learning in next-generation sequencing

“…Subsequently, a hybrid model named DanQ [40] was designed for that task and combines a CNN with an LSTM. More recently, DeepHistone [41] proposed a joint neural network module similar to the DeepCpG approach…”