From Raw Reads to Count Matrix: A Step-by-Step Guide to scATAC-seq Data Processing

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-20

Key Takeaways

The fundamental unit of scATAC-seq analysis is the DNA fragment, defined by the genomic coordinates of paired sequencing reads, which must be correctly reconstructed and assigned to a cell barcode.
Alignment of scATAC-seq reads requires specialized tools like Chromap that are significantly faster than traditional aligners (e.g., BWA-MEM) while maintaining comparable accuracy, crucial for handling large datasets.
Barcode processing is critical for cell calling, with methods like the ScSmOP pipeline employing spaced-seed hash table algorithms to accurately identify cell barcodes from both ligation-based and synthesis-based library preparations.
Fragment file generation consolidates aligned reads into fragments, filtering for proper pairs and quality, and serves as the input for downstream peak calling and count matrix construction.
Peak calling identifies accessible genomic regions, which then form the features for the count matrix, where each cell is quantified by the number of fragments overlapping these peaks.
Quality control metrics, including mapping rate, fragments per cell, and fraction of fragments in peaks, are essential at each stage to ensure the reliability of the final count matrix for downstream biological interpretation.

Single-cell ATAC-seq (scATAC-seq) generates genome-wide chromatin accessibility profiles from individual cells, but raw FASTQ files contain no direct information about which cell a read originated from or which genomic region was accessible. The core processing problem is converting these raw reads into a structured count matrix that links each cell to its chromatin accessibility profile across genomic regions. This guide walks through the complete workflow from raw sequencing data to a usable count matrix, covering alignment, barcode processing, fragment file generation, and the construction of peak-by-cell matrices. The focus is on practical decisions at each step, common failure points, and the quality metrics that determine whether downstream analysis will be reliable.

Understanding scATAC-seq Data Structure and Processing Requirements

What the Raw Data Contains

A typical scATAC-seq experiment produces FASTQ files containing sequencing reads from transposase-accessible chromatin fragments. Each read pair represents the two ends of a DNA fragment that was cut by the Tn5 transposase and amplified during library preparation. The critical feature of scATAC-seq libraries is that each fragment carries a cell-specific barcode sequence added during the tagmentation step. This barcode allows computational demultiplexing of pooled sequencing data into per-cell information.

The raw FASTQ files contain three essential components that must be processed correctly. First, the cell barcode sequence identifies the cell of origin for each fragment. Second, the genomic read sequences represent the actual DNA content from accessible chromatin regions. Third, the read pairing information allows reconstruction of the original DNA fragment, which is essential for understanding the insertion pattern of the Tn5 transposase.

The NCBI Data Resources provide access to public scATAC-seq datasets through the Sequence Read Archive and Gene Expression Omnibus, which are useful for testing processing pipelines before applying them to new experimental data. Understanding the structure of these public datasets helps researchers recognize the expected format of their own sequencing output.

Why scATAC-seq Processing Differs from scRNA-seq

The processing of scATAC-seq data presents distinct challenges compared to single-cell RNA-seq. The data are inherently sparse because each cell contains only two copies of the genome, and only a small fraction of the genome is accessible in any given cell type. This sparsity means that the count matrix will contain many zeros, and the interpretation of the data depends heavily on the quality of fragment processing.

Another key difference is that scATAC-seq reads are not directly mapped to genes. Instead, the reads map to genomic regions that may be distal regulatory elements such as enhancers or promoters. The count matrix is typically constructed at the level of peaks, which are regions of open chromatin identified across the dataset, instead of at the level of genes. This requires an additional peak-calling step that is not needed in standard scRNA-seq analysis.

The EMBL-EBI Training resources provide structured learning pathways for understanding the computational foundations of single-cell data analysis, including the specific considerations for chromatin accessibility data. Researchers new to this area should familiarize themselves with the concepts of genomic alignment, read counting, and sparse matrix operations before attempting to process scATAC-seq data.

Core Principles of scATAC-seq Data Processing

The Central Dogma of Fragment-Based Analysis

The fundamental unit of scATAC-seq analysis is the fragment, not the individual read. A fragment is defined by the genomic coordinates of the two read ends from a single sequencing read pair. The Tn5 transposase inserts into accessible chromatin and creates a characteristic pattern where the insertion sites are separated by approximately 9 base pairs on opposite strands. This pattern is used to identify true transposition events and to correct for the known sequence bias of the transposase.

The fragment file is the central intermediate product of scATAC-seq processing. It contains one line per fragment, specifying the cell barcode, the chromosome, the start position, the end position, and the number of reads supporting that fragment. This file format is used by most downstream analysis tools and serves as the input for constructing count matrices.

The Bioconductor project hosts many of the R packages used for scATAC-seq downstream analysis, including packages for reading fragment files, constructing count matrices, and performing quality control. The documentation for these packages provides detailed specifications of the fragment file format and the expected data structures.

The Role of Alignment in scATAC-seq Processing

Alignment is the process of determining where each sequencing read maps in the reference genome. For scATAC-seq, the alignment step must handle several complications that are less problematic in other assay types. The Tn5 transposase has a sequence bias that affects the probability of insertion at different genomic positions, and this bias must be accounted for during alignment and downstream analysis.

The alignment step also needs to handle reads that map to multiple locations in the genome. These multi-mapping reads are common in repetitive regions and can introduce noise if not handled properly. Most scATAC-seq pipelines discard multi-mapping reads or assign them to a single location using a probabilistic approach.

The Galaxy Training Network provides accessible tutorials for genomic alignment that cover the practical aspects of read mapping, including quality score handling, adapter trimming, and the interpretation of alignment statistics. These tutorials are useful for researchers who need to understand the alignment step in detail before applying it to their own data.

Practical Workflow from FASTQ to Count Matrix

Step 1: Quality Control of Raw Sequencing Reads

The first step in processing scATAC-seq data is assessing the quality of the raw FASTQ files. This involves checking the per-base quality scores, the GC content distribution, the presence of adapter sequences, and the overall read length distribution. Low-quality reads or reads with adapter contamination will produce poor alignment results and should be trimmed or filtered before proceeding.

The quality control step should also verify that the cell barcodes are present and correctly formatted. In most scATAC-seq protocols, the cell barcode is located in read 1 or read 2 of the paired-end reads, depending on the specific library preparation kit. The barcode length and position should match the expected format for the protocol being used.

The The Carpentries Lessons provide foundational training in shell scripting and data processing that is directly applicable to managing FASTQ files and running quality control tools. These lessons cover the command-line skills needed to efficiently process large sequencing datasets.

Step 2: Alignment to the Reference Genome

The alignment step maps the trimmed reads to the reference genome. For scATAC-seq data, the choice of aligner can significantly affect processing time and computational resource requirements. Traditional aligners such as BWA-MEM and Bowtie2 provide accurate results but can be slow for the deep sequencing depths typical of scATAC-seq experiments.

Faster alternatives have been developed specifically for chromatin profiling data. Chromap is an ultrafast method for aligning and preprocessing chromatin profiles that achieves alignment accuracy comparable to BWA-MEM and Bowtie2 while being over 10 times faster on bulk ChIP-seq and Hi-C profiles and faster than the standard Cell Ranger ATAC pipeline on single-cell ATAC-seq profiles. This speed improvement is achieved through efficient indexing and alignment strategies that exploit the specific characteristics of chromatin profiling data.

The alignment output should be sorted by genomic position and indexed to allow efficient random access during downstream processing. The alignment statistics should be reviewed to check the overall mapping rate, the proportion of reads mapping to the mitochondrial genome, and the fraction of reads that are properly paired.

Step 3: Barcode Processing and Cell Calling

After alignment, the cell barcodes must be extracted from the reads and used to assign each fragment to a cell. The barcode processing step involves correcting sequencing errors in the barcode sequences, filtering barcodes that do not match the expected whitelist, and identifying which barcodes correspond to real cells versus background or empty droplets.

The cell calling step is critical because it determines which barcodes will be included in the final count matrix. Barcodes with very few fragments may represent empty droplets or cells that were lost during library preparation. Barcodes with very high fragment counts may represent doublets where two cells were captured in the same droplet.

The ScSmOP pipeline demonstrates a universal approach to barcode identification that uses spaced-seed hash table-based algorithms to handle both ligation-based and synthesis-based barcoding data. This pipeline shows high reproducibility of data processing compared to published pipelines across multiple single-cell omics data types, including scATAC-seq.

Step 4: Fragment File Generation

The fragment file is generated by consolidating the aligned reads into fragments and assigning each fragment to a cell barcode. This step requires careful handling of read pairs to ensure that only properly paired reads are used to construct fragments. Reads that are not properly paired, that map to different chromosomes, or that have low mapping quality should be excluded.

The fragment file also records the number of reads supporting each fragment. This information is useful for quality control because fragments supported by multiple reads are more reliable than fragments supported by a single read. The Tn5 insertion pattern should also be checked at this stage to verify that the expected 9-base pair duplication pattern is present.

The Chromap method integrates alignment and preprocessing into a single step, generating fragment files directly from raw FASTQ data. This integrated approach reduces the computational overhead of writing and reading intermediate files and can significantly reduce the total processing time for large datasets.

Step 5: Peak Calling

Peak calling identifies the genomic regions that are consistently accessible across the cell population. These peaks represent candidate regulatory elements and serve as the features for the count matrix. Peak calling can be performed on the aggregated data from all cells or on subsets of cells defined by preliminary clustering.

The choice of peak calling method affects the number and width of peaks identified. Some methods identify narrow peaks that represent individual transcription factor binding sites, while others identify broader regions of accessibility. The peak width and the number of peaks should be appropriate for the biological question being addressed.

The pipeline described for single-cell chromatin accessibility analysis uses MACS2 for peak calling after initial data preprocessing with scATAC-pro or Cell Ranger ATAC. This approach identifies open chromatin regions and enables differential accessibility analysis to highlight regulatory differences among cell populations.

Step 6: Count Matrix Construction

The final step is constructing the count matrix that links each cell to the number of fragments overlapping each peak. This matrix is typically sparse because each cell only has accessibility at a small fraction of the identified peaks. The count matrix can be represented in several formats, including the standard sparse matrix format used by most single-cell analysis tools.

The count matrix construction should account for the fragment length distribution and the Tn5 insertion pattern. Some analysis methods use only the insertion sites instead of the full fragment, which can improve the resolution of the accessibility signal. The choice of whether to use full fragments or insertion sites depends on the downstream analysis goals.

The benchmarking study of computational methods for single-cell chromatin data analysis evaluated multiple feature engineering pipelines for their ability to discover and discriminate cell types. The study found that feature aggregation, SnapATAC, and SnapATAC2 outperformed latent semantic indexing-based methods, and that SnapATAC and SnapATAC2 were preferred for datasets with complex cell-type structures.

At a Glance: scATAC-seq Processing Pipeline Overview

Processing Stage	Primary Input	Key Output	Critical Quality Metric
Read quality control	Raw FASTQ files	Trimmed and filtered FASTQ	Per-base quality scores, adapter contamination rate
Alignment	Trimmed FASTQ	Sorted and indexed BAM	Mapping rate, mitochondrial read fraction
Barcode processing	Aligned reads	Cell barcode assignments	Number of cells identified, barcode error rate
Fragment generation	Aligned reads with barcodes	Fragment file	Fragment count per cell, Tn5 insertion pattern
Peak calling	Fragment file	Peak coordinates	Number of peaks, peak width distribution
Count matrix construction	Fragment file and peaks	Sparse count matrix	Matrix sparsity, fragments per cell in peaks

Tool Selection and Pipeline Options

Standard Pipelines for scATAC-seq Processing

Several complete pipelines are available for processing scATAC-seq data from raw reads to count matrix. The choice of pipeline depends on the computational resources available, the scale of the experiment, and the specific downstream analysis requirements.

The nf-core documentation describes community-developed pipelines that follow standardized best practices for reproducibility. These pipelines are designed to run on high-performance computing clusters and provide comprehensive reporting of quality metrics at each processing stage. The nf-core framework ensures that the processing steps are version-controlled and that the pipeline can be reproduced across different computing environments.

The ArchR framework provides an R-based bioinformatics framework that is well established for scATAC-seq datasets. This framework covers the complete analysis workflow from data preprocessing through downstream analysis, including quality control, clustering, and visualization. The ArchR framework is particularly useful for researchers who prefer to work within the R ecosystem.

Alignment-Free Approaches

Traditional scATAC-seq processing pipelines require substantial computational resources for alignment and downstream processing. Alignment-free approaches have been developed to reduce the computational burden while maintaining accuracy in downstream analysis.

One approach uses pseudoalignment to quantify reads against a reference set of genomic regions derived from DNase I Hypersensitive Sites. This method reduces execution times and hardware needs at little cost for precision. The approach was validated on public data for 10k PBMC and showed that cell groups identified are consistent with those identified from standard methods. Cell identification is robust when analysis is performed using DHS-derived reference in place of de novo identification of ATAC peaks.

The pseudoalignment approach demonstrates that analysis of scATAC-seq data by means of kallisto produces results in line with standard pipelines while being considerably faster. This approach is suitable for reliable quantification of gene activity based on scATAC-seq signal, allowing for efficient labeling of cell groups based on marker genes.

Deep Learning and Specialized Methods

Recent developments have introduced deep learning methods for specific aspects of scATAC-seq analysis. These methods address challenges in cell type annotation, data integration, and interpretation of chromatin accessibility patterns.

The scLLMDA framework uses a pretrained DNA-specific language model to generate contextual embeddings of peak sequences, which are then integrated with accessibility information to represent individual cells. This approach captures rich sequence semantics and neighborhood dependencies, enabling more accurate and robust cell type annotation across datasets.

The SEAGALL method uses geometry-regularized autoencoders and explainable graph attention networks to quantify the impact of molecular features on cellular phenotype. This method extracts specific and stable signatures from multiple omics experiments, going beyond differential marker genes to explain the features driving cell identity.

Quality Control Metrics and Thresholds

Fragment-Level Quality Metrics

The quality of scATAC-seq data is assessed at multiple levels, starting with individual fragments and progressing to the overall cell population. The fragment-level metrics include the number of reads supporting each fragment, the fragment length distribution, and the fraction of fragments that fall within called peaks.

The fragment length distribution is particularly informative for scATAC-seq quality assessment. Nucleosome-free fragments are typically shorter than 100 base pairs, while mononucleosome-associated fragments are approximately 180 to 247 base pairs. The ratio of nucleosome-free to nucleosome-associated fragments provides an indication of the signal-to-noise ratio in the data.

The transcription start site enrichment score measures the fraction of fragments that fall within a window around annotated transcription start sites. High-quality scATAC-seq data should show strong enrichment at transcription start sites because promoters are typically accessible in most cell types.

Cell-Level Quality Metrics

Cell-level quality metrics determine which barcodes are retained for downstream analysis. The total number of fragments per cell is a primary metric, with cells having very few fragments being excluded as likely empty droplets or low-quality cells. The fraction of fragments in peaks is another important metric, with high-quality cells typically having a substantial fraction of their fragments within called peaks.

The benchmarking study used 10 metrics calculated at the cell embedding, shared nearest neighbor graph, or partition levels to evaluate the performance of different feature engineering pipelines. This comprehensive approach allows for a thorough understanding of the strengths and weaknesses of each method and the influence of parameter selection.

Dataset-Level Quality Metrics

At the dataset level, the key metrics include the total number of cells identified, the median fragments per cell, the number of peaks called, and the fraction of reads that map to the mitochondrial genome. A high fraction of mitochondrial reads may indicate that cells are stressed or dying, which can affect the chromatin accessibility profile.

The MultiSC pipeline demonstrates how multi-omics data can be integrated to provide a more comprehensive understanding of cellular activities. For scATAC-seq data, the integration with gene expression and protein expression data can help validate the quality of the chromatin accessibility measurements.

Common Failure Patterns and Troubleshooting

Low Mapping Rate

A low mapping rate indicates that a substantial fraction of reads cannot be aligned to the reference genome. This can be caused by adapter contamination, sequencing errors, or contamination with non-genomic DNA. The solution depends on the cause, and the quality control step should identify the specific issue before proceeding with alignment.

If adapter contamination is the cause, the reads should be trimmed more aggressively before alignment. If sequencing errors are the cause, the quality filtering thresholds should be adjusted. If contamination is the cause, the source of contamination should be identified and addressed at the experimental level.

Poor Barcode Recovery

Poor barcode recovery results in fewer cells being identified than expected based on the number of nuclei loaded. This can be caused by barcode sequencing errors, incorrect barcode whitelist, or problems with the barcode design in the library preparation.

The ScSmOP pipeline demonstrates that barcode identification algorithms can be optimized for different barcoding strategies. The spaced-seed hash table-based algorithms used in this pipeline handle both ligation-based and synthesis-based barcoding data, showing that the choice of barcode identification algorithm can affect the number of cells recovered.

High Background Signal

High background signal is indicated by a low fraction of fragments in peaks and a high fraction of fragments in inaccessible regions. This can be caused by excessive tagmentation, incomplete nuclei isolation, or contamination with genomic DNA.

The pipeline for single-cell chromatin accessibility analysis includes quality control steps that can identify high background signal. The differential accessibility analysis can also help identify cells with unusual accessibility patterns that may indicate technical artifacts.

Batch Effects

Batch effects arise when samples are processed in different batches, introducing systematic differences that are not biologically meaningful. These effects can be particularly problematic in scATAC-seq data because the sparse and noisy nature of the data can amplify technical variation.

The CellSpace method provides a scalable and unbiased sequence-informed embedding of single-cell ATAC-seq data that can help address batch effects. The sequence-informed approach uses the genomic sequence context to improve the embedding, which can reduce the impact of technical variation.

Records and Documentation for Reproducibility

Version Control and Pipeline Documentation

Reproducibility in scATAC-seq processing requires careful documentation of the software versions, parameters, and reference files used at each step. The nf-core documentation emphasizes the importance of version control and provides guidelines for documenting pipeline usage and configuration.

The processing log should record the exact commands used, the software versions, the reference genome version, and the parameter values for each step. This information is essential for reproducing the analysis and for troubleshooting problems that may arise during downstream analysis.

Data Management and Storage

The intermediate files generated during scATAC-seq processing can be large, and proper data management is essential. The raw FASTQ files, the aligned BAM files, and the fragment files should be stored with appropriate backup and archival procedures.

The NCBI Data Resources provide repositories for depositing processed data, which is important for meeting data sharing requirements and for enabling other researchers to reproduce the analysis. The submission of processed data should include the count matrix, the peak coordinates, and the metadata describing the processing pipeline.

Analysis Notebooks and Reporting

The use of analysis notebooks that combine code, output, and narrative text can improve the reproducibility and transparency of scATAC-seq processing. The The Carpentries Lessons provide training in reproducible analysis practices, including the use of version control and automated testing.

The analysis report should include the quality metrics at each processing stage, the parameters used, and the decisions made during the processing. This report serves as a record of the analysis and provides context for interpreting the downstream results.

Integration with Downstream Analysis

From Count Matrix to Cell Types

The count matrix produced by the processing pipeline serves as the input for downstream analysis, including dimensionality reduction, clustering, and cell type annotation. The quality of the count matrix directly affects the reliability of these downstream analyses.

The benchmarking study provides guidelines for choosing analysis methods for different datasets. For datasets with complex cell-type structures, SnapATAC and SnapATAC2 are preferred. For large datasets, SnapATAC2 and ArchR are most scalable.

Multi-Omic Integration

Many experiments now generate scATAC-seq data alongside other modalities, such as gene expression or protein expression. The integration of these modalities requires careful alignment of the cell barcodes and the construction of joint count matrices.

The MultiSC pipeline demonstrates how multi-omics data can be integrated during the clustering process using a multimodal constraint autoencoder. This approach leverages the complementary information from different modalities to improve the identification of cell types and the prediction of gene regulatory networks.

The single-cell eQTL protocol describes how single-cell data can be integrated with genome-wide association study data to identify candidate risk genes and mechanisms underlying disease susceptibility. This integration requires careful processing of the single-cell data to ensure that the cell type annotations are accurate.

Trajectory and Regulatory Network Analysis

The count matrix can also be used for trajectory inference and regulatory network analysis. These analyses aim to reconstruct the developmental or activation trajectories of cells and to identify the transcription factors that drive these transitions.

The LAIOR framework provides a unified variational framework for interpretable single-cell manifold learning and trajectory inference. This framework combines Lorentz geometric regularization, a dual-path information bottleneck, and neural ODE regularization to preserve local cell-state structure, global hierarchy, and smooth developmental trajectories.

The pipeline for single-cell chromatin accessibility analysis includes transcription factor activity inference using chromVAR and the reconstruction of transcriptional regulatory networks using SCENIC+. These analyses provide insights into the regulatory mechanisms that control cell identity and function.

Limitations and Interpretation Considerations

Technical Limitations of scATAC-seq

The processing pipeline must account for the inherent technical limitations of scATAC-seq data. The data are sparse because each cell only provides a limited number of fragments, and the coverage of the genome is incomplete. This sparsity limits the resolution of the analysis and requires careful statistical handling.

The benchmarking study highlights the challenges in analyzing sparse, noisy, and high-dimensional data. The identification of cell types is a fundamental step in current single-cell data analysis practices, and the choice of feature engineering pipeline can significantly affect the results.

Biological Interpretation Limits

The count matrix represents chromatin accessibility, which is an indirect measure of regulatory activity. The presence of accessibility at a genomic region does not necessarily mean that the region is actively regulating gene expression. The interpretation of scATAC-seq data requires integration with other data types and careful consideration of the biological context.

The multi-omic single-cell landscape of perinatal mouse skin demonstrates how integrated single-cell chromatin and transcriptomic analyses can reveal key gene network axes underlying lineage specification. This study identified Mef2c+ upper fibroblasts as putative precursor cells for smooth muscle-like appendages, showing the power of multi-omic integration for biological discovery.

Computational Resource Considerations

The processing of scATAC-seq data requires substantial computational resources, particularly for large datasets. The alignment step is often the most resource-intensive, and the choice of aligner can significantly affect the processing time.

The pseudoalignment approach demonstrates that alignment-free techniques can reduce execution times and hardware needs at little cost for precision. This approach is particularly useful for researchers with limited computational resources.

Professional Escalation Criteria

When to Seek Expert Assistance

Certain situations warrant consultation with bioinformatics experts or the original equipment manufacturer. If the quality metrics indicate systematic problems that cannot be resolved through parameter adjustment, expert assistance should be sought.

Specific escalation criteria include: mapping rates below expected thresholds that persist after troubleshooting, barcode recovery that is substantially lower than expected, unexpected patterns in the fragment length distribution, and batch effects that cannot be corrected through standard normalization approaches.

Validation of Processing Results

Before proceeding with downstream analysis, the processing results should be validated using independent methods. This validation may include comparing the cell type annotations with known biological expectations, checking the reproducibility of the results across technical replicates, and verifying that the identified peaks are enriched for known regulatory elements.

The PerturbSeq.db repository provides a comprehensive collection of single-cell perturbation datasets that can be used for validation purposes. The uniform processing pipeline used in this repository ensures data consistency and comparability across datasets.

A Practical Decision Framework for Selecting scATAC-seq Processing Tools

Defining the Decision Context

Researchers processing scATAC-seq data face a crowded field of alignment tools, barcode processors, and peak callers, each with distinct performance characteristics. The choice of tool at each processing stage affects also runtime and memory usage but also the biological conclusions that can be drawn from the final count matrix. A structured decision framework helps researchers match tools to their specific experimental constraints instead of defaulting to the most popular option.

The benchmarking study of computational methods for single-cell chromatin data analysis evaluated eight feature engineering pipelines derived from five recent methods using ten metrics calculated at the cell embedding, shared nearest neighbor graph, or partition levels. The study found that feature aggregation, SnapATAC, and SnapATAC2 outperformed latent semantic indexing-based methods, and that SnapATAC and SnapATAC2 were preferred for datasets with complex cell-type structures. For large datasets, SnapATAC2 and ArchR were most scalable. These findings provide an evidence base for tool selection decisions.

Stage 1: Alignment Tool Selection Criteria

The alignment stage consumes the largest share of computational resources in scATAC-seq processing. The decision between traditional aligners and chromatin-specific aligners should be based on three primary criteria: dataset size, available computational infrastructure, and tolerance for alignment speed versus accuracy trade-offs.

Traditional aligners such as BWA-MEM and Bowtie2 provide accurate results but can become bottlenecks for the deep sequencing depths typical of scATAC-seq experiments. The Chromap method was designed specifically for chromatin profiling data and achieves alignment accuracy comparable to BWA-MEM and Bowtie2 while being over 10 times faster on bulk ChIP-seq and Hi-C profiles and faster than the standard Cell Ranger ATAC pipeline on single-cell ATAC-seq profiles. This speed improvement comes from efficient indexing and alignment strategies that exploit the specific characteristics of chromatin profiling data.

For datasets with fewer than 10,000 cells, the runtime difference between aligners may be negligible, and the choice can be based on familiarity and existing pipeline integration. For datasets exceeding 50,000 cells, the alignment time difference becomes substantial, and chromatin-specific aligners should be strongly considered. The decision should also account for whether the aligner produces output compatible with the downstream fragment processing tools being used.

Stage 2: Barcode Processing Strategy Selection

Barcode processing decisions depend on the library preparation method and the expected barcode structure. The ScSmOP pipeline demonstrates that barcode identification algorithms can be optimized for different barcoding strategies, using spaced-seed hash table-based algorithms for both ligation-based and synthesis-based barcoding data. This pipeline shows high reproducibility of data processing compared to published pipelines across multiple single-cell omics data types, including scATAC-seq.

The decision framework for barcode processing should consider whether the library preparation uses a fixed barcode whitelist or whether barcodes need to be identified de novo from the data. Fixed whitelists, such as those provided by commercial platforms, allow for straightforward error correction against known sequences. De novo barcode identification requires more sophisticated algorithms and may recover additional barcodes that are not in the expected whitelist.

Researchers should also consider whether their experiment uses a single-cell platform with known barcode structure or a custom protocol with unique barcode design. The ScSmOP pipeline handles both scenarios, making it a useful reference for understanding the range of barcode processing approaches available.

Stage 3: Peak Calling Parameter Decisions

Peak calling parameters determine the resolution and sensitivity of the regulatory element identification. The choice between narrow and broad peak calling affects the number of peaks identified and the interpretation of the count matrix. The pipeline described for single-cell chromatin accessibility analysis uses MACS2 for peak calling after initial data preprocessing with scATAC-pro or Cell Ranger ATAC, followed by differential accessibility analysis to detect open chromatin regions.

The decision framework for peak calling should consider the expected regulatory landscape of the cell types being studied. Cell types with well-characterized promoter-proximal regulatory elements may benefit from narrow peak calling that identifies individual transcription factor binding sites. Cell types with complex enhancer landscapes may require broader peak calling to capture the full extent of accessible regulatory regions.

The number of peaks identified affects the sparsity of the count matrix and the statistical power of downstream differential accessibility analysis. Too few peaks may miss important regulatory variation, while too many peaks may introduce noise and increase the multiple testing burden. The peak calling parameters should be validated by checking the fraction of fragments that fall within called peaks and the enrichment of known regulatory elements.

Stage 4: Count Matrix Resolution Decisions

The final count matrix can be constructed at different resolutions, including individual peaks, aggregated peak sets, or predefined genomic regions. The pseudoalignment approach demonstrates that using a reference set of DNase I Hypersensitive Sites instead of de novo identified peaks does not affect the ability to characterize cell populations. This finding supports the use of predefined genomic regions for count matrix construction when computational resources are limited.

The decision between de novo peak calling and predefined region quantification should consider the novelty of the biological system being studied. For well-characterized cell types with extensive public chromatin accessibility data, predefined regions may provide sufficient resolution. For poorly characterized cell types or novel biological contexts, de novo peak calling is necessary to capture the full range of accessible regulatory elements.

The count matrix resolution also affects the computational requirements of downstream analysis. Higher resolution matrices with more features require more memory and processing time for dimensionality reduction and clustering. The benchmarking study provides guidance on which analysis methods scale best with large datasets, with SnapATAC2 and ArchR being most scalable.

Implementing the Decision Framework

The decision framework should be implemented as a documented process that records the rationale for each tool selection. The nf-core documentation emphasizes the importance of version control and provides guidelines for documenting pipeline usage and configuration. Each processing decision should be recorded with the dataset characteristics, computational constraints, and biological considerations that informed the choice.

The implementation should include a validation step that compares the results from the selected tools against known biological expectations. This validation may include checking that the identified cell types match expected populations, that the peaks are enriched for known regulatory elements, and that the quality metrics meet the thresholds established for the specific experiment.

Common Decision Errors and Their Consequences

Several recurring errors undermine tool selection decisions in scATAC-seq processing. The first is selecting tools based on popularity instead of dataset-specific requirements. The benchmarking study demonstrates that different methods perform differently depending on the dataset characteristics, with SnapATAC and SnapATAC2 preferred for complex cell-type structures and SnapATAC2 and ArchR most scalable for large datasets.

The second common error is failing to account for the compatibility of tools across processing stages. Some aligners produce output formats that require additional conversion steps before barcode processing or fragment generation. These compatibility issues can introduce errors and increase processing time.

The third error is neglecting to validate tool performance on the specific dataset before committing to a full processing run. The Galaxy Training Network provides accessible tutorials for genomic alignment and processing that can be used to test tools on small subsets of data before scaling to the full dataset.

Recording Tool Selection Decisions

A tool selection record should document the dataset size, the computational infrastructure available, the biological question being addressed, and the rationale for each tool choice. This record serves as a reference for future experiments and for troubleshooting problems that may arise during downstream analysis.

The record should include the specific version numbers of all tools used, the parameter values for each processing step, and the quality metrics observed at each stage. The Bioconductor project hosts many of the R packages used for scATAC-seq downstream analysis, and the documentation for these packages provides detailed specifications of the expected input formats and parameter ranges.

The tool selection record should also note any deviations from the standard processing workflow and the reasons for those deviations. This documentation is essential for reproducing the analysis and for interpreting the biological results in the context of the processing decisions made.

Escalation Criteria for Tool Selection Problems

Certain situations warrant consultation with bioinformatics experts or the developers of the specific tools being used. If the selected tools produce inconsistent results across technical replicates, if the quality metrics indicate systematic problems that cannot be resolved through parameter adjustment, or if the computational requirements exceed the available infrastructure, expert assistance should be sought.

The EMBL-EBI Training resources provide structured learning pathways for understanding the computational foundations of single-cell data analysis, including the specific considerations for chromatin accessibility data. These resources can help researchers develop the expertise needed to make informed tool selection decisions and to troubleshoot problems when they arise.

The The Carpentries Lessons provide foundational training in shell scripting and data processing that is directly applicable to managing the computational aspects of scATAC-seq processing. These lessons cover the command-line skills needed to efficiently test and compare different processing tools.

Frequently Asked Questions

What is the minimum sequencing depth required for scATAC-seq?

The required sequencing depth depends on the biological question and the complexity of the cell population being studied. Higher depth provides more fragments per cell, which improves the resolution of the accessibility profile but increases the cost. The quality metrics at the cell level, such as the fraction of fragments in peaks, should be monitored to determine whether the depth is sufficient for the specific experiment.

How do I choose between different alignment tools for scATAC-seq data?

The choice of alignment tool depends on the computational resources available and the size of the dataset. Traditional aligners such as BWA-MEM and Bowtie2 provide accurate results but can be slow for large datasets. Faster alternatives such as Chromap provide comparable accuracy with significantly reduced processing time. The Chromap publication demonstrates that this method is over 10 times faster than traditional workflows on bulk chromatin profiles and faster than the standard Cell Ranger ATAC pipeline on single-cell ATAC-seq profiles.

What is the difference between a fragment file and a count matrix?

A fragment file contains one line per DNA fragment, specifying the cell barcode and the genomic coordinates of the fragment. A count matrix summarizes the fragment data by counting the number of fragments from each cell that overlap each genomic region of interest, such as a peak. The fragment file is an intermediate product that can be used to generate count matrices at different resolutions or for different sets of genomic regions.

How many cells should I expect to recover from my scATAC-seq experiment?

The number of cells recovered depends on the number of nuclei loaded, the capture efficiency of the platform, and the quality of the library preparation. The cell calling step identifies which barcodes correspond to real cells, and the number of cells identified should be consistent with the expected recovery rate for the specific platform and protocol being used.

What causes high mitochondrial read fraction in scATAC-seq data?

A high mitochondrial read fraction can indicate that cells are stressed or dying, which can affect the chromatin accessibility profile. The mitochondrial genome is more accessible in damaged cells, leading to a higher proportion of reads mapping to the mitochondrial genome. This metric should be monitored during quality control, and cells with unusually high mitochondrial read fractions may need to be excluded from downstream analysis.

Can I use the same processing pipeline for scATAC-seq and scRNA-seq data?

The processing pipelines for scATAC-seq and scRNA-seq differ in several important ways. scATAC-seq requires fragment-based processing and peak calling, while scRNA-seq requires transcript quantification. Some pipelines, such as ScSmOP, can handle both data types, but the specific processing steps and parameters differ between the assays.

How do I integrate scATAC-seq data with other single-cell modalities?

The integration of scATAC-seq data with other modalities requires careful alignment of the cell barcodes and the construction of joint count matrices. The MultiSC pipeline demonstrates how multi-omics data can be integrated during the clustering process, and the single-cell eQTL protocol describes how single-cell data can be integrated with genetic association data.

What are the most common mistakes in scATAC-seq data processing?

The most common mistakes include using incorrect barcode whitelists, failing to filter low-quality reads before alignment, using inappropriate peak calling parameters, and not monitoring quality metrics at each processing stage. The benchmarking study provides guidelines for choosing analysis methods and highlights the importance of parameter selection in determining the quality of the results.

Related Bioinformatics Guides

Related Clinical & Scientific Guides

References and Further Reading

NCBI Data Resources. National Center for Biotechnology Information.
EMBL-EBI Training. European Bioinformatics Institute.
Bioconductor. Bioconductor Project.
Galaxy Training Network. Galaxy Project.
nf-core Documentation. nf-core.
The Carpentries Lessons. The Carpentries.
Fast analysis of scATAC-seq data using a predefined set of genomic regions.. F1000Research, 2020.
MultiSC: a deep learning pipeline for analyzing multiomics single-cell data.. Briefings in bioinformatics, 2024.
ScSmOP: a universal computational pipeline for single-cell single-molecule multiomics data analysis.. Briefings in bioinformatics, 2023.
Benchmarking computational methods for single-cell chromatin data analysis.. Genome biology, 2024.
A pipeline for single-cell chromatin accessibility data analysis.. 2026.
Cell type annotation for scATAC-seq via DNA large language model and graph domain adaptation.. 2026.
Protocol for identifying cell-type-specific genes associated with disease risk using single-cell eQTL data.. 2026.
LAIOR: a hyperbolic neural ODE variational framework for interpretable single-cell manifold learning and trajectory inference.. 2026.
A multi-omic single-cell landscape of perinatal mouse skin maps lineage specification and reveals shared dynamics in human fetal skin.. 2026.
Geometry-aware graph attention networks to explain single-cell chromatin states and gene expression with SEAGALL.. 2026.
PerturbSeq.db: An integrated repository for comprehensive analysis of single-cell perturbation data.. Journal of Molecular Biology, 2025.
Fast alignment and preprocessing of chromatin profiles with Chromap. Nature Communications, 2021.
Using single-cell chromatin accessibility sequencing to characterize CD4+ T cells from murine tissues. Frontiers in Immunology, 2023.
Accelerating Single-Cell Sequencing Data Analysis with SciDAP: A User-Friendly Approach. Methods in Molecular Biology, 2025.
Scalable and unbiased sequence-informed embedding of single-cell ATAC-seq data with CellSpace. Nature Methods, 2024.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.