Predmoter - Cross-species prediction of plant promoter and enhancer regions

Abstract

MotivationThe identification of cis-regulatory elements (CREs) is crucial for the analysis of gene regulatory networks in plants. Several next generation sequencing (NGS)-based methods were developed to identify CREs. However, these methods can be time-consuming and costly. They also involve creating sequencing libraries for the entire genome. Since many research efforts only focus on specific genomic loci, this presents a considerable expenditure. Computational prediction of the outputs of specialized NGS methods to analyze CREs, like Assay for Transposase Accessible Chromatin using sequencing (ATAC-seq), would significantly cut costs and time investment. Yet, no such method is available to date. ResultsWe present Predmoter, a deep neural network able to predict base-wise ATAC-seq and histone Chromatin immunoprecipitation DNA-sequencing (ChIP-seq) read coverage for plant genomes. Predmoter uses only the DNA sequence as input. We evaluated our model on two plant genomes, the genome of the dicot Arabidopsis thaliana and of the monocot Oryza sativa. We trained our models on 10 species with publicly available ATAC-seq data and 15 species with ChIP-seq data. Our best models showed accurate predictions in peak positions and the overall pattern of peaks for ATAC- and Histone H3 trimethylated at lysine 4 (H3K4me3) ChIP-seq. Annotating putatively accessible chromatin regions provides valuable input for the identification of CREs. In conjunction with other in silico data, such as predicted binding affinities for transcription factors (TFs), this can significantly narrow down the search space to a manageable number of experimentally verifiable DNA-protein interaction pairs. Availability and ImplementationThe source code for Predmoter is available at: https://github.com/weberlab-hhu/Predmoter along with documentation for installation and usage. Predmoter uses a single-command inference, Predmoter.py, for both training and prediction. Predmoter takes a fasta file as input and outputs an h5 file and optionally bigWig and bedGraph files. HighlightPredmoter will help identifying CREs and so gaining further insight into gene regulatory networks in plants.