Pre-processing of 10X Single-Cell ATAC-seq Datasets

Overview
Creative Commons License: CC-BY Questions:
  • What is 10X?

  • What are single-cell ATAC fragments and MTX files?

  • What is an Anndata file, and why is it important?

Objectives:
  • Demultiplex single-cell FASTQ data from 10X Genomics

  • Learn about transparent matrix formats

  • Create a high-quality count matrix starting from scATAC-seq FASTQ files

Requirements:
Time estimation: 1 hour
Supporting Materials:
Published: Apr 24, 2023
Last modification: Jul 11, 2024
License: Tutorial Content is licensed under Creative Commons Attribution 4.0 International License. The GTN Framework is licensed under MIT
purl PURL: https://gxy.io/GTN:T00335
version Revision: 8

Similar to bulk ATAC-Seq, single-cell ATAC-Seq (scATAC-seq) leverages the hyperactive Tn5 Transposase to profile open chromatin regions but at single-cell resolution. Thus helps in understanding cell type-specific chromatin accessibility from a heterogeneous cell population.

Era of 10x Genomics

10x genomics has provided not only a cost-effective high-throughput solution to understanding sample heterogeneity at the individual cell level but has defined the standards of the field that many downstream analysis packages are now scrambling to accommodate. The gain in resolution reduces the granularity and noise issues that plagued the field of single-cell omics not long ago, where now individual clusters are much easier to decipher due to the added stability added by this gain in information.

Library Preparation

Library Preparation. Open image in new tab

Figure 1: An overview of the 10x single-nuclei ATAC-seq library preparation

The 10X barcoded gel beads consist of a pool of barcodes which are used to separately index each cell. The individual gel barcodes are delivered to each cell via flow cytometry, where each cell is fed single-file along a liquid tube and tagged with a 10X gel bead. The cells are then isolated from one another within thousands of nanoliter droplets, where each droplet is described by a unique 10x barcode that all reads in that droplet are associated with. The oil is then removed and all (now barcoded) DNA reads are pooled together to be sequenced.

More information can be found in the reagent kit documentation.

10x Chemistries

There are two main reagent kits used during the library preparation. Below we can see the layout of the primers used in both chemistries. We can ignore most of these as they are not relevant, namely: the P5 and P7 Illumina primers are used in the Illumina bridge amplification process; the Sample Index is an 8bp primer which is related to the Chromium system that balances nucleotide bias and ensures that there is no sample overlap during the multiplexed sequencing. The primer of interest to us is the Cell Barcode (CB). Unlike scRNA-seq, Unique Molecular Identifiers (UMIs) are not used in scATAC-seq.

chem. Open image in new tab

Figure 2: 10x scATAC-Seq sequencing library

There are two major chemistries (v1 and v2) used in 10x scATAC-seq library construction. Technically, v2 chemistry is more sensitive and has better data yield. But the final sequencing library is the same. Hence, same the data analysis applies to both chemistries. Both chemistries generate the 4 FASTQ files containing:

  • 8bp sample index sequence (I1)
  • 50bp forward read (R1)
  • 16bp barcode sequence (R2)
  • 49bp reverse read (R3)

The major differences to 10x scRNA-seq sequencing libraries are: (a) barcode sequences get a separate FASTQ file and (b) there are no UMIs. Note that as opposed to the traditional FASTQ naming, R3 contains reverse mate-pair and R2 contains barcode sequence.

Analysis Strategy

Agenda

In this tutorial, we will deal with:

  1. Era of 10x Genomics
  2. Library Preparation
  3. 10x Chemistries
  4. Analysis Strategy
  5. FASTQ processing and mapping
    1. Data upload and organization
    2. Attach cell barcodes to FASTQ headers
    3. FASTQ Quality control
    4. Mapping reads to a reference genome
  6. Peak calling
    1. Create scATAC-seq fragments file
    2. Call Peaks
  7. Count matrix creation
    1. AnnData
  8. Producing a Quality Count Matrix
    1. Initial filtering to remove potential empty features and cells
    2. Quality control and visualization
    3. Final filtering to acquire quality cells and features
  9. Conclusion

The tutorial is structured into four parts. The first part deals with the preprocessing of FASTQ files, quality control and mapping. Then in the second part, we will identify the open chromatin regions. In the third part, we will create a count matrix using the cell barcode and open chromatin regions information. In the final part, we will filter out empty cells and less informative peaks to produce a high-quality count matrix.

Here we will use a published data set from 10x genomics consisting of thousand peripheral blood mononuclear cells (PBMCs) extracted from a healthy donor. Analyzing these full datasets requires several hours of computation time. Hence, we subsampled the data so that the tutorial can be finished in a reasonable amount of time with sensible results. In the sub-sampled data, we have the reads that were mostly generated from chromosome 21. The original data has FASTQ files from two sequencing lanes L001 and L002. We merged the sequences from both lanes. We also ignored the sample index FASTQ file as it is not useful for this analysis. Finally, we are left with three FASTQ files: R1 (forward reads), R2 (barcodes) and R3 (reverse reads).

  • atac_pbmc_1k_nextgem_S1_R1_001_chr21.fastq.gz
  • atac_pbmc_1k_nextgem_S1_R2_001_chr21.fastq.gz
  • atac_pbmc_1k_nextgem_S1_R3_001_chr21.fastq.gz

These files are provided in the Zenodo data repository. To analyze the full datasets, get the original FASTQ files from 10x genomics website

FASTQ processing and mapping

Data upload and organization

We import all three FASTQ files into a history.

Hands-on: Data upload and organization
  1. Create a new history and rename it (e.g. scATAC-seq 10X preprocessing tutorial)

    To create a new history simply click the new-history icon at the top of the history panel:

    UI for creating new history

  2. Import the sub-sampled FASTQ data from Zenodo or from the data library (ask your instructor)

    As an alternative to uploading the data from a URL or your computer, the files may also have been made available from a shared data library:

    1. Go into Data (top panel) then Data libraries
    2. Navigate to the correct folder as indicated by your instructor.
      • On most Galaxies tutorial data will be provided in a folder named GTN - Material –> Topic Name -> Tutorial Name.
    3. Select the desired files
    4. Click on Add to History galaxy-dropdown near the top and select as Datasets from the dropdown menu
    5. In the pop-up window, choose

      • “Select history”: the history you want to import the data to (or create a new one)
    6. Click on Import

    https://zenodo.org/record/7855968/files/atac_pbmc_1k_nextgem_S1_R1_001_chr21.fastq.gz
    https://zenodo.org/record/7855968/files/atac_pbmc_1k_nextgem_S1_R2_001_chr21.fastq.gz
    https://zenodo.org/record/7855968/files/atac_pbmc_1k_nextgem_S1_R3_001_chr21.fastq.gz
    
    • Copy the link location
    • Click galaxy-upload Upload Data at the top of the tool panel

    • Select galaxy-wf-edit Paste/Fetch Data
    • Paste the link(s) into the text field

    • Press Start

    • Close the window

Question

Inspect all the uploaded FASTQ files by clicking on the galaxy-eye symbol.

  1. Which FASTQ file contains the barcode sequence?
  2. What is the length of the barcode?
  1. In this case, R2 contains the cell barcodes
  2. 16bp

Attach cell barcodes to FASTQ headers

First, we will attach the barcodes from the barcodes FASTQ file to the ids of forward and reverse reads that carry the association between reads and barcodes throughout the analysis. We will use this information to demultiplex the cells at the end while creating the count matrix. We will be left with only two FASTQ files of forward and reverse reads after this step.

Comment

Sinto barcode ( Galaxy version 0.9.0+galaxy1) consumes the barcode sequence from the atac_pbmc_1k_nextgem_S1_R2_001_chr21.fastq.gz file and uses : delimiter to prepend it to both the forward and reverse read ids. The tool expects the same number of reads in all three FASTQ files and should be present in the same order.

Hands-on

Sinto barcode ( Galaxy version 0.9.0+galaxy1) with the following parameters:

  • “FASTQ file containing cell barcode sequences”: atac_pbmc_1k_nextgem_S1_R2_001_chr21.fastq.gz
  • “Single or Paired-end data”: Paired
  • “Forward reads FASTQ file”: atac_pbmc_1k_nextgem_S1_R1_001_chr21.fastq.gz
  • “Reverse reads FASTQ file”: atac_pbmc_1k_nextgem_S1_R3_001_chr21.fastq.gz
  • “Number of bases to extract from barcode-containing FASTQ”: 16
Comment: Sinto barcode output

After a successful run, the barcode sequences from the barcode FASTQ file are prepended to the read names. We can inspect the output FASTQ files by clicking on the galaxy-eye symbol of the barcoded read 1 file.

Now let’s proceed with the next steps QC, mapping and peak calling as we do with the bulk ATAC-Seq data. For a thorough description of bulk ATAC-Seq analysis, please follow the bulk ATAC-Seq data analysis tutorial. Here we use a slightly different workflow. Here we map reads using BWA-MEM instead of Bowtie2, then create ATAC fragments file that also stores the cell barcode information and finally use MACS2 for peak calling. Note that the workflow from bulk ATAC-seq tutorial should also result in similar set of peaks.

FASTQ Quality control

First things first. Let’s do a basic FASTQ quality control using FastQC

Hands-on
  1. FastQC ( Galaxy version 0.73+galaxy0) with the following parameters:
    • “Raw read data from your current history”: Use param-files Multiple datasets to choose both barcoded read 1 and barcoded read 2 (outputs of Sinto barcode tool).
  2. Inspect the web page output of FastQC tool for the barcoded read 1 sample.
Question
  1. How many reads are in the FASTQ?
  2. Which sections have a warning?
  1. There are 685686 reads.
  2. The 2 steps below have warnings:

    1. Per base sequence content

      It is well known that the Tn5 has a strong sequence bias at the insertion site. You can read more about it in Green et al. 2012.

    2. Sequence Duplication Levels

      The read library quite often has PCR duplicates that are introduced simply by the PCR itself. We will remove these duplicates later on.

There is only a little amount of Nextera Transposase Sequence in the reads (see Adapter Content section of FastQC output). In this case, we skip the adapter trimming step and proceed to the mapping step. If your data has a significant amount of Nextera Transposase Sequences, then proceed to adapter trimming using either cutadapt or Trim Galore!. Trim Galore can automatically detect the adapter sequence. If you use Cutadapt, follow the bulk ATAC-Seq tutorial Trimming Reads step.

Mapping reads to a reference genome

Now we map the reads to a reference genome using BWA-MEM ( Galaxy version 0.7.17.2).

Hands-on: Mapping

Map with BWA-MEM ( Galaxy version 0.7.17.2) with the following parameters:

  • “Using reference genome”: hg38
  • In “Single or Paired-end reads”: Paired
    • In “Select first set of reads “: barcoded read 1 (output of Sinto barcode tool)
    • In “Select second set of reads “: barcoded read 2 (output of Sinto barcode tool)
  • “Select analysis mode”: 1.Simple Illumina mode
  • “BAM sorting mode”: Sort by chromosomal coordinates

Peak calling

In scRNA-seq we always have the standard set of genes (usually downloaded from public databases) to quantify the expression levels. For scATAC-seq, there are no such reference open chromatin regions because the regions of chromatin accessibility are tissue dependent. Hence, we first have to detect open chromatin regions. There are several ways of doing this. The most common ways are the following:

  • Find a published bulk ATAC-seq data that is the closest to your scATAC-seq data and use the regions from the bulk ATAC-seq data
  • Chunk the genome into equal-sized bins and use these bins as reference locations for quantification
  • Combine the data from all the cells of the scATAC-seq data together and detect open chromatin regions from the data. Later use these detected regions for quantification

In this tutorial, we opt for the 3rd option.

Create scATAC-seq fragments file

An ATAC-seq fragment file is a BED file with Tn5 integration sites, the cell barcode associated with the fragment, and the frequency of the sequenced fragment. PCR duplicates are collapsed. Here we filter out the alignments with low mapping quality.

Comment: on Tn5 insertion

When Tn5 cuts an accessible chromatin locus it inserts adapters separated by 9bp (Kia et al. 2017):

Nextera_Library_Construction. Open image in new tab

Figure 3: Nextera Library Construction

This means, to have the read start site reflecting the center of where Tn5 bound, the reads on the positive strand should be shifted 4 bp to the right and reads on the negative strands should be shifted 5 bp to the left as in Buenrostro et al. 2013. Genrich can apply these shifts when ATAC-seq mode is selected. In most cases, we do not have 9bp resolution so we don’t take it into account but if you are interested in the footprint, this is important.

Hands-on: Create scATAC fragments
  1. Sinto fragments ( Galaxy version 0.9.0+galaxy1) with the following parameters:
    • “Input BAM file”: mapped reads in BAM format (output of Map with BWA-MEM tool)`
    • “Minimum MAPQ required to retain fragment”: 30
    • “Regular expression used to extract cell barcode from read name”: [^:]* (matches all characters up to the first colon)
    • “Number of bases to shift Tn5 insertion position by on the forward strand”: 4
    • “Number of bases to shift Tn5 insertion position by on the reverse strand”: -5
  2. bedtools SortBED ( Galaxy version 2.30.0+galaxy2) with the following parameters:
    • “Sort the following BED/bedGraph/GFF/VCF/EncodePeak file *“: fragments BED (output of Sinto fragments tool)`
    • “Sort by”: chromosome, then by start position (asc)
  3. Rename the dataset sorted fragments

Call Peaks

We have now finished the data preprocessing. Next, to find regions corresponding to potential open chromatin regions, we want to identify regions where reads have piled up (peaks) greater than the background read coverage. The tools which are currently used are Genrich and MACS2. MACS2 is more widely used. At this step, two approaches exist:

  • The first one is to select only pairs whose fragment length is below 100bp corresponding to nucleosome-free regions and to use a peak calling like you would do for a ChIP-seq, joining signal between mates. The disadvantage of this approach is that you can only use it if you have paired-end data and you will miss small open regions where only one Tn5 is bound.
  • The second one chosen here is to use all reads to be more exhaustive. In this approach, it is very important to re-center the signal of each read on the 5’ extremity (read start site) as this is where Tn5 cuts. Indeed, you want your peaks around the nucleosomes and not directly on the nucleosome:
ATAC-Seq_reads_relative_to_nucleosomes. Open image in new tab

Figure 4: Scheme of ATAC-Seq reads relative to nucleosomes

If we only assess the coverage of the 5’ extremity of the reads, the data would be too sparse and it would be impossible to call peaks. Thus, we will extend the start sites of the reads by 200bp (100bp in each direction) to assess coverage. We call peaks with MACS2. In order to get the coverage centered on the 5’ extended 100bp on each side we will use --shift -100 and --extend 200:

MACS2_peak_centering_options. Open image in new tab

Figure 5: MACS2 options to get 100bp each side
Hands-on: Call peaks with MACS2
  1. MACS2 callpeak ( Galaxy version 2.1.1.20160309.6) with the following parameters:
    • “Are you pooling Treatment Files?”: No
    • “ChIP-Seq Treatment File”: sorted fragments
    • “Do you have a Control File?”: No
    • “Format of Input Files”: Single-end BED
    • “Effective genome size”: H. sapiens (2.7e9)
    • “Build Model”: Do not build the shifting model (--nomodel)
      • “Set extension size”: 200
      • “Set shift size”: -100. It needs to be - half the extension size to be centered on the 5’.
    • “Additional Outputs”:
      • Check Peaks as tabular file (compatible with MultiQC)
      • Check Peak summits
      • Check Scores in bedGraph files
    • In “Advanced Options”:
      • “Composite broad regions”: No broad regions
        • “Use a more sophisticated signal processing approach to find subpeak summits in each enriched peak region”: Yes
      • “How many duplicate tags at the exact same location are allowed?”: all
    Comment: Why keeping all duplicates is important

    We previously removed duplicates using Sinto fragment tool using paired-end information. If two pairs had identical R1 but different R2, we knew it was not a PCR duplicate. Because we converted the BAM to BED we lost the pair information. If we keep the default (removing duplicates) one of the 2 identical R1 would be filtered out as duplicate.

    Comment: BED / encode narrowPeak format

    If you are not familiar with BED format or encode narrowPeak format, see the BED Format

Count matrix creation

Now we have ATAC fragments with cell barcode information as well as the peaks to create a scATAC count matrix. In the context of ATAC-seq, the peaks are also often called features. But the term feature is a generic term that denotes a peak in the case of scATAC-seq data and a gene in case of scRNA-seq data. So the count matrix we build is similar to that of scRNA-seq data with the exception that here we have open chromatin regions instead of genes. In this tutorial, both the terms feature and peak indicate an open chromatin region.

For count matrix creation, we will use Build count matrix from EpiScanpy tool suite. It takes fragments file and the peaks in either BED format or directly a narrowPeak file of MACS2 output. Regardless of file type, the tool only considers the first 3 columns which contain chromosome id, start and end positions of the peak regions. Parameters like --call-summits from our MACS2 run can result in different peaks with the same peak boundaries. It is always a good practice to remove duplicate peak regions so that our final count matrix will have unique features.

Hands-on: Extract unique peak regions
  1. Unique occurrences of each record ( Galaxy version 1.1.0) with the following parameters:
    • “File to scan for unique values”: narrow Peaks (output of MACS2 tool)
    • In “Advanced Options”: Show Advanced Options
      • “Column start”: Column: 1
      • “Column end”: Column: 3
Question
  1. How many peak regions were duplicated?
  1. There were initially 1064 regions in the narrow Peaks file. Now there are 891 regions after deduplication. More than 15% (173) of regions have the same peak boundaries.

AnnData

The most common format, called AnnData, stores the matrix as well as gene and cell annotations in a concise, compressed and extremely readable manner:

Anndata format. Open image in new tab

Figure 6: AnnData format stores a count matrix X together with annotations of observations (i.e. cells) obs, variables (i.e. peaks) var and unstructured annotations uns.

This format is used by Scanpy (Wolf et al. 2018) and EpiScanpy (Danese et al. 2021) tool suites for analyzing single-cell omics data. So we need first to import the matrix and annotations of peaks and cells (present in fragments BED file) into an AnnData object.

Hands-on: Build count matrix with EpiScanpy
  1. Build count matrix with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “ATAC fragments file”: sorted fragments
    • “Features file”: output of Unique tool tool
    • “Normalize peak sizes?”: Keep the peaks as they are
    • In “Select the chromosomes of the species you are considering”: Custom list of chromosomes
      • “Enter comma seperated list of chromosome ids (without chr prefix)”: 21
  2. Rename the generated file to PBMC 1k chr21 Anndata
  3. Check that the format is h5ad

In this example data, our reads are from the chr21. Hence, we input 21 for chromosome ids. For human or mouse genomes, directly select from the dropdown list. Make sure that all the canonical chromosomes are present in the fragments file.

Because the AnnData format is an extension of the HDF5 format, i.e. a binary format, an AnnData object can not be inspected directly in Galaxy by clicking on the galaxy-eye (View data) icon. Instead, we need to use a dedicated tool from the AnnData suite.

Hands-on: Inspect an AnnData object
  1. Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
    • param-file “Annotated data matrix”: PBMC 1k chr21 Anndata
    • “What to inspect?”: General information about the object
  2. Inspect the generated file

    Question
    AnnData object with n_obs × n_vars = 18426 × 891
    
    1. How many observations are there? What do they represent?
    2. How many variables are there? What do they represent?
    1. There are 18,426 observations, representing the cells.
    2. There are 891 variables, representing the peaks.
    Comment: Faster Method for General Information
    • To view general information of any Anndata file:
      • Click on the name of the dataset in the history to expand it.
      • General Anndata information would be given in the expanded box:

        e.g.

        [n_obs x n_vars]
        -    18426 x 891
        
    • For more specific queries, Inspect AnnData ( Galaxy version 0.7.5+galaxy1) is required.
  3. Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
    • param-file “Annotated data matrix”: PBMC 1k chr21 Anndata
    • “What to inspect?”: The full data matrix
  4. Inspect the generated file

    Question
    	chr21_10272276_10272566	chr21_10330743_10330987	chr21_10489276_10489476	chr21_13979698_13980574	chr21_14027127_14027558 ...
    AAAAAATACCCGAAGG	0.0	0.0	0.0	0.0
    AAAAACAAAAATATAC	0.0	0.0	0.0	0.0
    AAAAATAACAATGATA	0.0	0.0	0.0	0.0
    AAAACCGTGAGAGTCT	0.0	0.0	0.0	0.0
    AAAACCTCACCCAACA	0.0	0.0	0.0	0.0
    

    What is stored in the generated file?

    The file is a table with 18,426 lines (observations or cells) and 891 columns (variables or peaks): the count matrix for each of the 891 peaks and 18,426 cells. The 1st row contains the peak location as an annotation of the columns and the 1st column the barcodes of the cells as an annotation of the rows.

  5. Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
    • param-file “Annotated data matrix”: PBMC 1k chr21 Anndata
    • “What to inspect?”: Key-indexed observations annotation (obs)
  6. Inspect the generated file

    Question
    AAAAAATACCCGAAGG
    AAAAACAAAAATATAC
    AAAAATAACAATGATA
    AAAACCGTGAGAGTCT
    AAAACCTCACCCAACA
    

    What is stored in the generated file?

    The file is tabular with annotations of the observations. i.e. the cells. Here there are only the barcodes as annotation, so only one column, being also the index for the count matrix.

  7. Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
    • param-file “Annotated data matrix”: PBMC 1k chr21 Anndata
    • “What to inspect?”: Key-indexed annotation of variables/features (var)
  8. Inspect the generated file

    Question
    chr21_10272276_10272566
    chr21_10330743_10330987
    chr21_10489276_10489476
    chr21_13979698_13980574
    chr21_14027127_14027558
    

    What is stored in the generated file?

    The file is tabular with annotations of the variables, i.e. the peaks.

Producing a Quality Count Matrix

We just created a count matrix with all detected peaks and non-empty barcodes. In this section of the tutorial, we will filter the features and barcodes by a series of quality assessments to generate a high-quality count matrix. The final count matrix can be used for further downstream analysis like clustering, differential chromatin analysis and cell type annotation.

The current matrix contains peaks from chromosome 21 only and is too small to carry out this type of analysis. From this step onwards we will use the full count matrix generated from the original FASTQ files. Please import the following anndata file to start:

https://zenodo.org/record/7855968/files/atac_pbmc_1k_uniq_peaks.h5ad
Question
  1. How many barcodes and features are there?
  2. What do you observe in comparison to a scRNA-seq count matrix?
  1. There are 441038 barcodes and 70120 features
  2. The initial number of barcodes in this raw count matrix is in the same range as the scRNA-seq data. But there is roughly double the number of features compared to a scRNA-seq dataset. The number of features varies from dataset to dataset. Based on the tools and parameters used for peak calling, the size of the initial feature set can grow up to several hundred of thousands.

Initial filtering to remove potential empty features and cells

First remove any potential empty features or barcodes. A non-empty cell should have a minimum number of non-empty features and a non-empty feature should be present in a minimum number of non-empty cells. Here we will different minimum thresholds for filtering cells and features.

Hands-on: Initial filtering
  1. scATAC-seq Preprocessing with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “Annotated data matrix”: atac_pbmc_1k_uniq_peaks.h5ad
    • “Method used for filtering”: Filter cell outliers based on counts and numbers of features expressed, using 'pp.filter_cells'
    • “Filter”: Minimum number of features expressed
    • “Minimum features”: 100
  2. scATAC-seq Preprocessing with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “Annotated data matrix”: Output of the previous (pp.filter_cells) step
    • “Method used for filtering”: Filter features based on counts and numbers of features expressed, using 'pp.filter_features'
    • “Filter”: Minimum number of cells expressed
    • “Minimum features”: 1
Question

How much data was filtered out?

The resulting matrix has dimensions of 1815 x 67766, i.e., more than 99.5% of the cells and less than 4% of features were filtered out. This indicates the high sparsity of the count matrix.

Quality control and visualization

We will first plot the number of features per cell. This information is stored as nb_features layer of observations. As we are dealing with numbers with huge variances, it is sometimes hard to visualize them. Hence, we compute the log of nb_features first then plot the absolute and log values together.

Hands-on: Violing plots of number of features
  1. scATAC-seq Preprocessing with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “Annotated data matrix”“: Output of previous (pp.filter_features) step
    • “Method used for filtering”: Compute log10 of nb_features
  2. Check if observations have an extra log_nb_features in addition to nb_features layer.
  3. Rename the resulting dataset to PBMC 1k after initial filtering
  4. Plot with scanpy ( Galaxy version 1.7.1+galaxy0) with the following parameters:
    • param-file “Annotated data matrix”: PBMC 1k after initial filtering
    • “Method used for plotting”: Generic: Violin plot, using 'pl.violin'
      • “Keys for accessing variables”: Subset of variables in 'adata.var_names' or fields of '.obs'
        • “Keys for accessing variables”: nb_features, log_nb_features
      • In “Violin plot attributes”:
        • “Add a stripplot on top of the violin plot”: Yes
          • “Add a jitter to the stripplot”: Yes
            • “Size of the jitter points”: 0.4
        • “Display keys in multiple panels”: Yes
      • In “Parameters for seaborn.violinplot”:
        • “Color for all of the elements”: DarkCyan
Question

What trends of the data do you observe in the plots?

IntialViolin. Open image in new tab

Figure 7: Violin plot of number of features per cell after initial filtering

Both plots indicate that there are two distinct sets of cells:

  1. Cells with an acceptable number of open chromatin regions.
  2. Cells with only a few detected peaks. These cells are uninformative and should be excluded.

To determine decent filtering thresholds, we will further look at some histograms of cell-peak coverage trends. We will plot histograms of the number of open features per cell and feature commonness in cells.

Hands-on: Coverage cells
  1. scATAC-seq Preprocessing with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “Annotated data matrix”“: PBMC 1k after initial filtering
    • “Method used for filtering”: Coverage cells: Histogram of the number of open features (in the case of ATAC-seq data) per cell, using 'pp.coverage_cells'
    • “Binarized matrix?”: No
    • “Minimum number of cells or minimum number of features to be indicated in the plot”: 1000
  2. scATAC-seq Preprocessing with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “Annotated data matrix”“: PBMC 1k after initial filtering
    • “Method used for filtering”: Coverage cells: Histogram of the number of open features (in the case of ATAC-seq data) per cell, using 'pp.coverage_cells'
    • “Binarized matrix?”: No
    • “Log transform?”: Yes
    • “Minimum number of cells or minimum number of features to be indicated in the plot”: 1000
  3. Now observe the plots
Question
  1. What information do you see in these plots?
  2. Do you think our minimum number of features threshold of 1000 indicated in the plot makes sense?
CoverageCells. Open image in new tab

Figure 8: Histogram of the number of open chromatin regions per cell
CoverageCellsLog. Open image in new tab

Figure 9: Histogram of the number of open chromatin regions per cell in log scale
  1. The plots show the number of open chromatin regions per cell in 50 bins on the x-axis and the number of cells per bin on the y-axis. It is basically a histogram of the number of open chromatin regions per cell.
  2. From the first plot with absolute counts, we see that with our threshold of 1000, we can separate between the cells with few detected peaks and cells with high chromatin accessibility.
Hands-on: Coverage features
  1. scATAC-seq Preprocessing with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “Annotated data matrix”“: PBMC 1k after initial filtering
    • “Method used for filtering”: Coverage features: Distribution of the feature commonness in cells, using 'pp.coverage_features'
    • “Binarized matrix?”: No
    • “Minimum number of cells or minimum number of features to be indicated in the plot”: 5
  2. scATAC-seq Preprocessing with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “Annotated data matrix”“: PBMC 1k after initial filtering
    • “Method used for filtering”: Coverage features: Distribution of the feature commonness in cells, using 'pp.coverage_features'
    • “Binarized matrix?”: No
    • “Log transform?”: Yes
    • “Minimum number of cells or minimum number of features to be indicated in the plot”: 5
  3. Now observe the plots
Question
  1. What information do you see in these plots?
  2. Do you think our minimum number of cells threshold of 5 indicated in the plot makes sense?
CoverageFeatures. Open image in new tab

Figure 10: Histogram of feature commonness in cells
CoverageFeaturesLog. Open image in new tab

Figure 11: Histogram of feature commonness in cells in log scale
  1. The plots show a histogram of the number of cells sharing a feature. As we initially pooled the data from all the cells to detect the peaks, it is expected to see only a small number of cells have more than 10000 peaks in common.
  2. The red vertical line of our 5 cells threshold is nearly at the left end of the histogram representing the majority of the features have at least 5 cells in common. From the log scale plot it is also clear that there is a sharp increase in the feature commonness from at least 10 cells (x-axis 1.0). So our threshold of 5 is a decent cutoff for filtering out the features. From the plots, only a very few non-informative features are left to be filtered out.

Final filtering to acquire quality cells and features

Based on the above QC plots, we will filter out all the cells with less than 1000 features and all the features that have less than 5 cells in common. Additionally, we will also filter out extreme outlier cells (which can be seen in the initial violin plot or coverage cells plot) with more than 12500 features.

Hands-on: Final filtering based on QC plots
  1. scATAC-seq Preprocessing with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “Annotated data matrix”: PBMC 1k after initial filtering
    • “Method used for filtering”: Filter cell outliers based on counts and numbers of features expressed, using 'pp.filter_cells'
    • “Filter”: Minimum number of features expressed
    • “Minimum features”: 1000
  2. scATAC-seq Preprocessing with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “Annotated data matrix”: Output of previous (pp.filter_cells) step
    • “Method used for filtering”: Filter cell outliers based on counts and numbers of features expressed, using 'pp.filter_cells'
    • “Filter”: Maximum number of features expressed
    • “Minimum features”: 12500
  3. scATAC-seq Preprocessing with EpiScanpy ( Galaxy version 0.3.2+galaxy1) with the following parameters:
    • “Annotated data matrix”: Output of previous (pp.filter_cells) step
    • “Method used for filtering”: Filter features based on counts and numbers of features expressed, using 'pp.filter_features'
    • “Filter”: Minimum number of cells expressed
    • “Minimum features”: 5
Question

What are the dimensions of the count matrix after the final filtering?

The resulting matrix has 1024 cells and 67719 peaks. The number of cells after filtering is close to what we expect in this dataset.

Conclusion

In this tutorial, we have learned how to map reads from 10X ATAC-seq data, identify open chromatin regions from aggregated cell data and generate a count matrix in Anndata format. We also performed quality control and filtered out empty and uninformative cells and features to generate a high-quality count matrix for downstream analysis.

Galaxy workflows for generating Annadata from FASTQ files and quality filtering of raw count matrix are provided.scatac_tenx.ga).

This tutorial is part of the https://singlecell.usegalaxy.eu portal (Tekman et al. 2020).