Assessment of computational methods for the analysis of single-cell ATAC-seq data

Abstract

Background: Recent innovations in single-cell Assay for Transposase Accessible Chromatin using sequencing (scATAC-seq) enable profiling of the epigenetic landscape of thousands of individual cells. scATAC-seq data analysis presents unique methodological challenges. scATAC-seq experiments sample DNA, which, due to low copy numbers (diploid in humans), lead to inherent data sparsity (1-10% of peaks detected per cell) compared to transcriptomic (scRNA-seq) data (10-45% of expressed genes detected per cell). Such challenges in data generation emphasize the need for informative features to assess cell heterogeneity at the chromatin level. Results: We present a benchmarking framework that is applied to 10 computational methods for scATAC-seq on 13 synthetic and real datasets from different assays, profiling cell types from diverse tissues and organisms. Methods for processing and featurizing scATAC-seq data were compared by their ability to discriminate cell types when combined with common unsupervised clustering approaches. We rank evaluated methods and discuss computational challenges associated with scATAC-seq analysis including inherently sparse data, determination of features, peak calling, the effects of sequencing coverage and noise, and clustering performance. Running times and memory requirements are also discussed. Conclusions: This reference summary of scATAC-seq methods offers recommendations for best practices with consideration for both the non-expert user and the methods developer. Despite variation across methods and datasets, SnapATAC, Cusanovich2018, and cisTopic outperform other methods in separating cell populations of different coverages and noise levels in both synthetic and real datasets. Notably, SnapATAC is the only method able to analyze a large dataset (> 80,000 cells).

From reads to insight: a hitchhiker’s guide to ATAC-seq data analysis

“…Despite the analyses listed for bulk ATAC-seq, another important analysis for single-cell is clustering. A recent benchmarking study from Chen et al about clustering methods in scATAC-seq showed that SnapATAC, Cusanovich2018 and cisTopic outperformed other methods [23,[173][174][175]. These three methods are featured by workflows combining window-based genome binning, binarization of the accessibility, coverage bias correction, and dimension reduction using principle component analysis, which specifically handle the sparse scATAC-seq data [175].…”

Tuning parameters of dimensionality reduction methods for single-cell RNA-seq analysis

“…type identification as a downstream task of DR. Both measures are frequently used to assess the performance of cell embedding techniques [18,11,20,23,10]. Another standard measure to assess the ability of a clustering algorithm to recover known classes is the adjusted rand index (ARI) [11,20,23,10]; however, we found that AMI and ARI are extremely correlated (Additional file 1: Fig.…”

“…For Louvain, the nearest neighbors graph construction was done with scanpy (version 1.3.8) using default parameters, and the clustering was also run with default parameters using the 'taynaud' flavor. Note that since we had to automatically run Louvain on all embeddings, as done in [23], we could not properly tune the resolution nor the size of the neighborhood and thus Louvain could either overcluster or undercluster.…”

scATAC-pro: a comprehensive workbench for single-cell chromatin accessibility sequencing data

“…We generated simulated data with different levels of noise from the bulk ATAC-seq data of 13 primary human blood cell types [8] using the same strategy as that in [47]. We started with the bulk peak-by-cell count matrix and generated count for peak i in cell type t using a binomial distribution binomð2; p t i Þ, where p t i ¼ ð1−qÞr t i =2 þ qn=2k, r t i is the percentage of all reads overlapping with peak i in cell type t, k is the total number of peaks in the bulk data, n is the number of simulated fragments, and q is a parameter specifying the level of noise; q = 0 indicates no noise while q = 1 indicates the highest level of noise.…”