4 Construction of expression matrix

Many analyses of scRNA-seq data take as their starting point an expression matrix. By convention, the each row of the expression matrix represents a gene and each column represents a cell (although some authors use the transpose). Each entry represents the expression level of a particular gene in a given cell. The units by which the expression is meassured depends on the protocol and the normalization strategy used.

4.1 Reads QC

The output from a scRNA-seq experiment is a large collection of cDNA reads. The first step is to ensure that the reads are of high quality. The quality control can be performed by using standard tools, such as FastQC or Kraken.

Assuming that our reads are in experiment.bam, we run FastQC as

$<path_to_fastQC>/fastQC experiment.bam

Below is an example of the output from FastQC for a dataset of 125 bp reads. The plot reveals a technical error which resulted in a couple of bases failing to be read correctly in the centre of the read. However, since the rest of the read was of high quality this error will most likely have a negligible effect on mapping efficiency.

Example of FastQC output

Figure 4.1: Example of FastQC output

Additionally, it is often helpful to visualize the data using the Integrative Genomics Browser (IGV) or SeqMonk.

4.2 Reads alignment

After trimming low quality bases from the reads, the remaining sequences can be mapped to a reference genome. Again, there is no need for a special purpose method for this, so we can use the STAR or the TopHat aligner. For large full-transcript datasets from well annotated organisms (e.g. mouse, human) pseudo-alignment methods (e.g. Kallisto, Salmon) may out-perform conventional alignment. For drop-seq based datasets with tens- or hundreds of thousands of reads pseudoaligners become more appealing since their run-time can be several orders of magnitude less than traditional aligners.

An example of how to map reads.bam to using STAR is

$<path_to_STAR>/STAR --runThreadN 1 --runMode alignReads
--readFilesIn reads1.fq.gz reads2.fq.gz --readFilesCommand zcat --genomeDir <path>
--parametersFiles FileOfMoreParameters.txt --outFileNamePrefix <outpath>/output

Note, if the spike-ins are used, the reference sequence should be augmented with the DNA sequence of the spike-in molecules prior to mapping.

Note, when UMIs are used, their barcodes should be removed from the read sequence. A common practice is to add the barcode to the read name.

Once the reads for each cell have been mapped to the reference genome, we need to make sure that a sufficient number of reads from each cell could be mapped to the reference genome. In our experience, the fraction of mappable reads for mouse or human cells is 60-70%. However, this result may vary depending on protocol, read length and settings for the read alignment. As a general rule, we expect all cells to have a similar fraction of mapped reads, so any outliers should be inspected and possibly removed. A low proportion of mappable reads usually indicates contamination.

An example of how to quantify expression using Salmon is

$<path_to_Salmon>/salmon quant -i salmon_transcript_index -1 reads1.fq.gz -2 reads2.fq.gz -p #threads -l A -g genome.gtf --seqBias --gcBias --posBias

Note Salmon produces estimated read counts and estimated transcripts per million (tpm) in our experience the latter over corrects the expression of long genes for scRNASeq, thus we recommend using read counts.

4.3 Alignment example

The histogram below shows the total number of reads mapped to each cell for an scRNA-seq experiment. Each bar represents one cell, and they have been sorted in ascending order by the total number of reads per cell. The three red arrows indicate cells that are outliers in terms of their coverage and they should be removed from further analysis. The two yellow arrows point to cells with a surprisingly large number of unmapped reads. In this example we kept the cells during the alignment QC step, but they were later removed during cell QC due to a high proportion of ribosomal RNA reads.

Example of the total number of reads mapped to each cell.

Figure 4.2: Example of the total number of reads mapped to each cell.

4.4 Mapping QC

After mapping the raw sequencing to the genome we need to evaluate the quality of the mapping. There are many ways to measure the mapping quality, including: amount of reads mapping to rRNA/tRNAs, proportion of uniquely mapping reads, reads mapping across splice junctions, read depth along the transcripts. Methods developed for bulk RNA-seq, such as RSeQC, are applicable to single-cell data:

python <RSeQCpath>/geneBody_coverage.py -i input.bam -r genome.bed -o output.txt
python <RSeQCpath>/bam_stat.py -i input.bam -r genome.bed -o output.txt
python <RSeQCpath>/split_bam.py -i input.bam -r rRNAmask.bed -o output.txt

However the expected results will depend on the experimental protocol, e.g. many scRNA-seq methods use poly-A selection to avoid sequencing rRNAs which results in a 3’ bias in the read coverage across the genes (aka gene body coverage). The figure below shows this 3’ bias as well as three cells which were outliers and removed from the dataset:

Example of the 3' bias in the read coverage.

Figure 4.3: Example of the 3’ bias in the read coverage.

4.5 Reads quantification

The next step is to quantify the expression level of each gene for each cell. For mRNA data, we can use one of the tools which has been developed for bulk RNA-seq data, e.g. HT-seq or FeatureCounts

# include multimapping
<featureCounts_path>/featureCounts -O -M -Q 30 -p -a genome.gtf -o outputfile input.bam
# exclude multimapping
<featureCounts_path>/featureCounts -Q 30 -p -a genome.gtf -o outputfile input.bam

Unique molecular identifiers (UMIs) make it possible to count the absolute number of molecules and they have proven popular for scRNA-seq. We will discuss how UMIs can be processed in the next chapter.

4.6 Unique Molecular Identifiers (UMIs)

Thanks to Andreas Buness from EMBL Monterotondo for collaboration on this section.

4.6.1 Introduction

Unique Molecular Identifiers are short (4-10bp) random barcodes added to transcripts during reverse-transcription. They enable sequencing reads to be assigned to individual transcript molecules and thus the removal of amplification noise and biases from scRNASeq data.

UMI sequencing protocol

Figure 4.4: UMI sequencing protocol

When sequencing UMI containing data, techniques are used to specifically sequence only the end of the transcript containing the UMI (usually the 3’ end).

4.6.2 Mapping Barcodes

Since the number of unique barcodes (\(4^N\), where \(N\) is the length of UMI) is much smaller than the total number of molecules per cell (~\(10^6\)), each barcode will typically be assigned to multiple transcripts. Hence, to identify unique molecules both barcode and mapping location (transcript) must be used. The first step is to map UMI reads, for which we recommend using STAR since it is fast and outputs good quality BAM-alignments. Moreover, mapping locations can be useful for eg. identifying poorly-annotated 3’ UTRs of transcripts.

UMI-sequencing typically consists of paired-end reads where one read from each pair captures the cell and UMI barcodes while the other read consists of exonic sequence from the transcript (Figure 4.5). Note that trimming and/or filtering to remove reads containing poly-A sequence is recommended to avoid erors due to these read mapping to genes/transcripts with internal poly-A/poly-T sequences.

After processing the reads from a UMI experiment, the following conventions are often used:

  1. The UMI is added to the read name of the other paired read.

  2. Reads are sorted into separate files by cell barcode
    • For extremely large, shallow datasets, the cell barcode may be added to the read name as well to reduce the number of files.
UMI sequencing reads, red lightning bolts represent different fragmentation locations

Figure 4.5: UMI sequencing reads, red lightning bolts represent different fragmentation locations

4.6.3 Counting Barcodes

In theory, every unique UMI-transcript pair should represent all reads originating from a single RNA molecule. However, in practice this is frequently not the case and the most common reasons are:

  1. Different UMI does not necessarily mean different molecule
    • Due to PCR or sequencing errors, base-pair substitution events can result in new UMI sequences. Longer UMIs give more opportunity for errors to arise and based on estimates from cell barcodes we expect 7-10% of 10bp UMIs to contain at least one error. If not corrected for, this type of error will result in an overestimate of the number of transcripts.
  2. Different transcript does not necessarily mean different molecule
    • Mapping errors and/or multimapping reads may result in some UMIs being assigned to the wrong gene/transcript. This type of error will also result in an overestimate of the number of transcripts.
  3. Same UMI does not necessarily mean same molecule
    • Biases in UMI frequency and short UMIs can result in the same UMI being attached to different mRNA molecules from the same gene. Thus, the number of transcripts may be underestimated.
Potential Errors in UMIs

Figure 4.6: Potential Errors in UMIs

4.6.4 Correcting for Errors

How to best account for errors in UMIs remains an active area of research. The best approaches that we are aware of for resolving the issues mentioned above are:

  1. UMI-tools’ directional-adjacency method implements a procedure which considers both the number of mismatches and the relative frequency of similar UMIs to identify likely PCR/sequencing errors.

  2. Currently an open question. The problem may be mitigated by removing UMIs with few reads to support their association with a particular transcript, or by removing all multi-mapping reads.

  3. Simple saturation (aka “collision probability”) correction proposed by Grun, Kester and van Oudenaarden (2014) to estimate the true number of molecules \(M\):

\[M \approx -N*log(1 - \frac{n}{N})\] where N = total number of unique UMI barcodes and n = number of observed barcodes.

An important caveat of this method is that it assumes that all UMIs are equally frequent. In most cases this is incorrect, since there is often a bias related to the GC content.

Per gene amplification rate

Figure 4.7: Per gene amplification rate

Determining how to best process and use UMIs is currently an active area of research in the bioinformatics community. We are aware of several methods that have recently been developed, including:

4.6.5 Downstream Analysis

Current UMI platforms (DropSeq, InDrop, ICell8) exhibit low and highly variable capture efficiency as shown in the figure below.

Variability in Capture Efficiency

Figure 4.8: Variability in Capture Efficiency

This variability can introduce strong biases and it needs to be considered in downstream analysis. Recent analyses often pool cells/genes together based on cell-type or biological pathway to increase the power. Robust statistical analyses of this data is still an open research question and it remains to be determined how to best adjust for biases.

Exercise 1 We have provided you with UMI counts and read counts from induced pluripotent stem cells generated from three different individuals (Tung et al. 2017) (see: Chapter 7.1 for details of this dataset).

umi_counts <- read.table("tung/molecules.txt", sep = "\t")
read_counts <- read.table("tung/reads.txt", sep = "\t")

Using this data:

  1. Plot the variability in capture efficiency

  2. Determine the amplification rate: average number of reads per UMI.


Tung, Po-Yuan, John D. Blischak, Chiaowen Joyce Hsiao, David A. Knowles, Jonathan E. Burnett, Jonathan K. Pritchard, and Yoav Gilad. 2017. “Batch Effects and the Effective Design of Single-Cell Gene Expression Studies.” Sci. Rep. 7 (January). Springer Nature: 39921. doi:10.1038/srep39921.