A comprehensive comparison of differential accessibility analysis methods for ATAC-seq data

Background: ATAC-seq is widely used to measure the chromatin accessibility and identify the open chromatin regions (OCRs). OCRs usually indicate the active regulatory elements in the genome and are directly associated with gene regulatory networks. Identification of differential accessibility regions (DARs) between different biological conditions is critical to measure the differential activity of regulatory elements. Differential analysis of ATAC-seq shares many similarities to differential expression analysis of RNA-seq data. However, the distribution of ATAC-seq signal is different from RNA-seq data, and higher sensitivity is desired for DARs identification. Many different tools can be used to perform differential analysis of ATAC-seq data, but a comprehensive comparison and benchmarking of these methods is still missing. Methods: Here, we used simulated datasets to systematically measure the sensitivity and specificity of 6 different methods. We further discussed the statistical and signal density cutoff in the differential analysis of ATAC-seq by applying to real data. Batch-effect is very common in high-throughput sequencing experiments. Results: We illustrated that batch-effect correction can dramatically improve the sensitivity in differential analysis of ATAC-seq data. Finally, we developed an easily usable package, BeCorrect, to perform batch-effort correction for visualizing corrected ATAC-seq signals on a genome browser. Conclusions: It is important to use PCA to check the samples distribution, and the Remove Unwanted Variation strategy can be used to correct the data to improve the sensitivity when strong batch effects are found in the data. Finally, BeCorrect can be used to correct the batch-effect of ATAC-seq data signal based on DARs analysis, and generate a proper visualization on a genome browser. This represents the dentate gyrus and Ammon’s horn samples only