sgdemux

This repository is home to the sgdemux tool for demultiplexing sequencing data generated on Singular Genomics’ sequencing instruments.

• Installation
  • From Bioconda
  • From Releases
  • From Source
• Contributing
• Overview
• Usage
  • Inputs
    • FASTQ Files
      • Auto-detecting FASTQS from a Path Prefix
    • Read Structures
    • Specifying Sample Information
      • Sample Sheet
      • Simple Two-column CSV
    • Full Argument List
    • Performance Considerations
  • Outputs
    • Demultiplexed FASTQs
    • Metrics
      • per_sample_metrics.tsv
      • per_project_metrics.tsv
      • metrics.tsv
      • most_frequent_unmatched.tsv
      • sample_barcode_hop_metrics.tsv
• Advance Usage
  • Single Sample

Installation

sgdemux may be installed from bioconda, downloaded from the releases page, or built from source.

From Bioconda

Install from bioconda with:

conda create -n sgdemux -c bioconda sgdemux
conda activate sgdemux

From Releases

Install from pre-built binaries on the Releases page

From Source

1. Install rust and cargo
2. Install dependencies. For example, cmake and build-essentials are required for Ubuntu 22.04. Install using commands below.

sudo apt-get update
sudo apt-get install build-essential cmake -y

Note: cmake for older OS version such as Ubuntu 18.04 is not incompatible.

3. Clone the repo and build the software:

git clone https://github.com/Singular-Genomics/singular-demux.git
cd singular-demux
cargo install --path ../singular-demux --locked

Contributing

Contributions are welcome. See the Contributing Guidelines for details.

Overview

sgdemux performs sample demultiplexing on block-compressed (BGZF) FASTQs such as those produced by the Singular Genomics G4 platform. The input FASTQs must be block compressed (e.g. with bgzip); uncompressed or non-bgzf gzipped input files are not supported as performance would be significantly degraded.

The primary options that affect demultiplexing are --allowed-mismatches and --min-delta. Together these specify a) how well a sample barcode in a sequencing read must match an expected barcode and b) how much worse the next best match must be. The default options of --allowed-mismatches 1 --min-delta 2 will only match a set of FASTQ records to an expected barcode if, across all barcode reads, there is at most one mismatch (allowed mismatches) vs. the expected barcode and the difference (minimmum delta) between the number of mismatches of the best and second best matching barcode is greater than two mismatches. Note: the allowed mismatches is not used when determining the next-best matching barcode.

For additional examples, consider --allowed-mismatches 3 --min-delta 1, with two barcodes b1 and b2:

1. If b1 matches with 2 mismatches, and b2 matches with 3 mismatches, then the delta between the number of mismatches is 1, which is not greater than --min-delta, and therefore the read is not assigned to a barcode.
2. If b1 matches with 1 mismatch, and b2 matches with 3 mismatches, then the delta between the number of mismatches is 2, which is greater than --min-delta, and therefore the read is assigned to barcode b1.
3. If b1 matches with 0 mismatch, and b2 matches with 2 mismatches, then the delta between the number of mismatches is 2, which is greater than --min-delta, and therefore the read is assigned to barcode b1.
4. If b1 matches with 0 mismatches, and b2 matches with 1 mismatches, then the delta between the number of mismatches is 1, which is not greater than --min-delta, and therefore the read is not assigned to a barcode.
5. If b1 matches with 3 mismatches, and b2 matches with 6 mismatches, then the delta between the number of mismatches is 3, which is greater than --min-delta, and the number of mismatches for b1 is less than equal to than --allowed-mismatches, and thefore read is assigned to barcode b1.
6. If b1 matches with 4 mismatches, and b2 matches with 6 mismatches, then the delta between the number of mismatches is 2, which is greater than --min-delta, but the number of mismatches for b1 is greater than --allowed-mismatches, and thefore the read is not assigned to a barcode.
7. If b1 matches with 2 mismatch, and b2 matches with 2 mismatches, then the delta between the number of mismatches is 0, which is not greater than --min-delta, and therefore the read is not assigned to a barcode.

Several other options affect how demultiplexing is performed, and for these to be fully understood it is necessary to understand the order in which they are applied in the demultiplexing process. Operations are ordered as follows:

1. A record is read in from each of the input FASTQ files and broken into read “segments” using the supplied read structures.
2. If --filter-control-reads is specified and the reads are marked as controls in the FASTQ header, the reads are discarded (i.e. they do not get written to any output files).
3. If --filter-failing-quality is specified and the reads are marked as quality failures in the FASTQ header, the reads are discarded (i.e. they do not get written to any output files).
4. If one or more --quality-mask-threshold values are supplied, template bases in all input reads that have base quality below the given threshold value are masked to N.
5. Match the reads against the set of expected barcodes; if the sample barcode has more N bases in it that specified by --max-no-calls or does not match to an expected barcode within defined parameters, the reads will be assigned to the undetermined sample.
6. Write out the subset of the FASTQs/read segments specified by --output-types to the FASTQ file(s) for the assigned sample.

Usage

The primary inputs to the tool are:

1. A set of undemultiplexed FASTQ files (BGZF compressed)
2. A set of read-structures, one per input FASTQ file
3. A file of sample metadata including sample names and barcode sequences
4. A directory into which the demultiplexed FASTQ files should be written

Reads are written to per-sample, per-instrument-read files within the output directory. An additional Undetermined set of files will be written containing those reads that did not match any expected barcodes.

An example invocation follows:

sgdemux \
  --fastqs R1.fastq.gz R2.fastq.gz I1.fastq.gz I2.fastq.gz \
  --read-structures +T +T 8B 8B \
  --sample-metadata sample-metadata.csv \
  --output-dir demuxed/

Inputs

FASTQ Files

The full set of FASTQ files generated for a run, or lane, or sequencing should be provided, including all template and index reads. For example if a paired-end sequencing run was performed with dual sample index reads, four files should be provided:

  --fastqs R1.fastq.gz R2.fastq.gz I1.fastq.gz I2.fastq.gz

If multiple FASTQ files are available per instrument reads, they should be concatenated prior to running sgdemux. BGZF files, due to their block-compressed nature, can be concatenated simply using standard cat, e.g.:

for read in R1 R2 I1 I2; do cat L*/${read}.fastq.gz > ./${read}.fastq.gz; done

FASTQ files must be BGZF compressed.

Auto-detecting FASTQS from a Path Prefix

Alternatively, the FASTQS can be auto-detected when a path prefix is given to --fastqs <dir>/<prefix>. The FASTQs must be named <dir>/<prefix>_L00<lane>_<kind><kind-number>_001.fastq.gz, where kind is one of R (read/template), I (index/sample barcode), or U (umi/molecular barcode).

The Read Structure will be derived from file names (kind and kind number), with the full read length used for the given kind. The derived Read Structure and FASTQs will be ordered first by kind (I then R then U), second by read number (e.g. R1 before R2). This is important for command line options that can be specified once per read kind and number. E.g. if the following FASTQs are present with path prefix /path/to/prefix:

/path/to/prefix_L001_I1_001.fastq.gz
/path/to/prefix_L001_I2_001.fastq.gz
/path/to/prefix_L001_R1_001.fastq.gz
/path/to/prefix_L001_R2_001.fastq.gz

then the +B +B +T +T read structure will be used. Since this tool requires all sample barcode segments to have a fixed length, the first read in any index/sample-barcode FASTQ will be examined and its length used as the expected sample barcode length.

Furthermore, multiple lanes may be given and will be used for demultiplexing:

/path/to/prefix_L001_I1_001.fastq.gz
/path/to/prefix_L002_I1_001.fastq.gz
/path/to/prefix_L001_I2_001.fastq.gz
/path/to/prefix_L002_I2_001.fastq.gz
/path/to/prefix_L001_R1_001.fastq.gz
/path/to/prefix_L002_R1_001.fastq.gz
/path/to/prefix_L001_R2_001.fastq.gz
/path/to/prefix_L002_R2_001.fastq.gz

When data for multiple lanes is provided, each lane must have the same number and types of input fastqs.

The auto-detected/derived Read Structure may be overriden on the command line or in the sample sheet by providing the --read-structures argument. In this case, the new read structure must be given and will be applied in the same order as described above (e.g. I1, I2, R1, R2 for a dual index paired end run).

Read Structures

Read Structures are short strings that describe the origin and/or purpose of bases within sequencing reads. They are made up of a sequence of <number><operator> pairs (segments). Four kinds of operators are recognized:

1. T identifies template reads/bases
2. B identifies sample barcode reads/bases
3. M identifies unique molecular index reads/bases
4. S identifies a set of bases to be skipped or ignored

The last <number><operator> pair in a Read Structure may use + instead of a number to denote “all remaining bases”. This is useful if, e.g., FASTQs have been trimmed and/or contain reads of varying length.

For more details on Read Structures, and how to validate them, see this detailed description.

Read Structures are not required to be provided when using a path prefix for the input FASTQs. In that case, the read structure will be inferred from the FASTQ name. See: Auto-detecting FASTQS from a Path Prefix.

When providing the input FASTQs explicitly, one Read Structure must be provided for each input FASTQ file, in the same order. Matching the set of reads specified in the FASTQ files section above one might specify:

  --read-structures +T +T 8B 8B

All sample barcode segments must be a fixed length. E.g. 8B+T is allowed but 10S+B is not.

Specifying Sample Information

The sample metadata file may be a Sample Sheet or a simple two-column CSV file with headers.

Sample Sheet

Information about the sample(s) to demultiplex is specified within a Sample Sheet. Command line options for demultiplexing may also be passed via the Sample Sheet.

The Sample Sheet may have a [Demux] section for command line options, and must have a [Data] section for sample information.

The [Demux] section must contain a line per command line option. The first column must contain the option long name without the leading -- (e.g. fastqs or read-structures). The second column contains the option value, or empty if the option takes no value (i.e. a flag). If the option accepts multiple values, they must be space separated in the second column. The command line options specified in the sample sheet override those provided on the command line. The order of the FASTQs must match the order read structures.

The [Data] section must contain a header line. The Sample_ID column must contain a unique, non-empty identifier for each sample. One or both of Index1_Sequence and Index2_Sequence must be present with values for indexed runs. For non-indexed runs, a single sample must be given with an empty value for both the Index1_Sequence and Index2_Sequence columns. Both Sample_IDs and the Index1_Sequence/Index2_Sequence combinations must be unique within the file, and both columns are required for all samples.

An example follows:

[Demux]
fastqs,/path/to/i1.fq.gz /path/to/r1.fq.gz
read-structures,+B +T
[Data]
Sample_ID,Index1_Sequence,Index2_Sequence
s1,ACTGGTCA,
s2,ATACGAAC,

Simple Two-column CSV

For the simple two-column CSV, the Sample_Barcode column must contain the unique set of sample barcode bases for the sample(s). If multiple sample barcodes are are present (e.g. dual-indexing runs, additional inline sample indices) then the Sample_Barcode field should contain the full set of barcode bases expected to be read for the sample. The ordering of the concatenated barcodes is important, and should match the ordering of the FASTQs and Read Structures given. Both Sample_IDs and Sample_Barcodes must be unique within the file, and both columns are required for all samples. An example follows:

Sample_ID,Sample_Barcode
s1,ACTGGTCA
s2,ATACGAAC

For example if a dual-indexing run was performed with an additional inline sample barcode in read 1, and sgdemux was invoked with the following options:

--fastqs R1.fastq.gz I1.fastq.gz I2.fastq.gz R2.fastq.gz \
--read-structures 10B+T 8B 8B +T

then the Sample_Barcode field for each sample should be composed as follows:

  {10 base inline index}-{8 base I1 index}-{8 base I2 index}

Full Argument List

Argument Name	Required	Default Value	Description
–fastqs	Yes	n/a	Path(s) to the input FASTQs, or path prefix if not a file.
–sample-metadata	Yes	n/a	Path to CSV of sample metadata with sample IDs and barcode sequences.
–read-structures	No	n/a	Read structures, one per input FASTQ. Do not provide when using a path prefix for FASTQs.
–output-dir	Yes	n/a	Path to an output directory to write into.
–allowed-mismatches	No	1	The number of mismatches allowed, in total, between expected and observed barcode bases in order to match a read to a sample.
–min-delta	No	2	The minimum number of mismatches by which the best match for a read is better than the next-best match for a read in order to accept the best match.
–free-ns	No	1	The number of observed Ns (no-calls) in the barcode read(s) that are allowed for “free” before treating subsequent Ns as mismatches.
–max-no-calls	No	n/a	If specified, do not match any reads whose barcode reads contain more than this many Ns.
–quality-mask-threshold	No	n/a	Mask to N template bases in all input reads whose base quality is below the specified value(s). A single value may be specified, which is then applied to all input reads/FASTQs. Alternatively one value per input FASTQ may be provided in the same order as the FASTQs. Sample barcode/index and UMI bases are never masked.
–filter-control-reads	No	False	If true, filter out reads marked as control reads in their FASTQ headers.
–filter-failing-quality	No	False	If true, filter out reads marked as failing quality control in their FASTQ headers.
–output-types	No	T	The types of bases/reads for which output files should be generated. A single string containing one or more of T (template), B (sample barcode), M (UMI), and S (skipped).
–undetermined-sample-name	No	Undetermined	The name used as a prefix to generate FASTQ files for reads that didn’t match to any sample.
–most-unmatched-to-output	No	1000	Report on the top N most frequently observed unmatched barcode sequences.
–demux-threads	No	4	The number of threads to use to perform demultiplexing in memory.
–compressor-threads	No	12	The number of threads to use in compressing the output FASTQ files.
–writer-threads	No	5	The number of threads to use to write compressed FASTQ data to disk.
–override-matcher	No	n/a	The algorithm to use for matching, either CachedHammingDistance or PreCompute. By default if barcodes are 12bp or shorter PreCompute is used which pre-computes all possible matches, or if barcodes are longer than 12bp CachedHammingDistance is used which calculates matches when needed then caches the results.
–skip-read-name-check	No	False	If this is true, then all the read names across FASTQs will not be enforced to be the same. This may be useful when the read names are known to be the same and performance matters. Regardless, the first read name in each FASTQ will always be checked.
–sample-barcode-in-fastq-header	No	False	If this is true, then the sample barcode is expected to be in the FASTQ read header. For dual indexed data, the barcodes must be + (plus) delimited. Additionally, if true, then neither index FASTQ files nor sample barcode segments in the read structure may be specified.
–metric-prefix	No	n/a	Prepend this prefix to all output metric file names.
–lane	No	n/a	Select a subset of lanes to demultiplex. Will cause only samples and input FASTQs with the given Lane(s) to be demultiplexed. Samples without a lane will be ignored, and FASTQs without lane information will be ignored.

Performance Considerations

Various --*-threads options are available to control the number of threads used by sgdemux for various purposes. The defaults are intended to fully utilize a 32-core machine. The defaults to the available options do not add up to 32 as several threads are used to read the input FASTQ files and for ancillary purposes.

For running on larger or smaller instances it is advised to start with the following and tune from there:

• 1/3 of available threads for compression
• 1/6 of available threads for writing
• 1/6-1/3 of available threads for demultiplexing

Currently this tool does not provide a way place a hard limit on the number of threads used.

Outputs

Demultiplexed FASTQs

One or more BGZF compressed FASTQ files will be created per sample in the specified output directory. For paired end data, the output will have the suffix _R1.fastq.gz and _R2.fastq.gz for read one and read two respectively.

Samples barcodes, and unique molecular indices (UMIs), will be inserted into the FASTQ headers if present. If either multiple sample barcodes or multiple UMIs are present they will be concatenated with + between individual barcodes prior to insertion. For example if a FASTQ record had sample barcodes ACGT and TTGA, and UMIs of ACCTAG and TCCTGG the the output header might look like:

sg001:17:A30ZZ:1:44567:ACCTAG+TCCTGG 1:N:1:ACGT+TTGA

Metrics

Up to five metrics files are generated to help assess run and demultiplexing quality:

per_sample_metrics.tsv

This file always produced and contains the following columns:

Column	Description
sample_ID	The name for the sample barcode, typically the same name from the SampleSheet.
barcode	The sample barcode bases. Dual index barcodes will have two sample barcode sequences delimited by a +.
total_matches	The total number of templates matching the given barcode.
perfect_matches	The number of templates that match perfectly the given barcode.
one_mismatch_matches	The number of pass-filter templates that match the given barcode with exactly one mismatch.
q20_bases	The number of bases in a template with a quality score 20 or above.
q30_bases	The number of bases in a template with a quality score 30 or above.
total_number_of_bases	The total number of bases in the templates combined.
fraction_matches	The fraction of all templates that match the given barcode.
ratio_this_barcode_to_best_barcode	The ratio of templates for this barcode to the number of templates of the most prevelant barcode (excluding Undetermined).
frac_q20_bases	The fraction of bases in a template with a quality score 20 or above.
frac_q30_bases	The fraction of bases in a template with a quality score 30 or above.
mean_index_base_quality	The mean quality of index bases.

The per_sample_metrics.tsv file produces a row per sample.

per_project_metrics.tsv

The per_project_metrics.tsv file aggregates the metrics by project (aggregates the metrics across samples with the same project) and has the same columns as per_sample_metrics.tsv. In this case, sample_ID will contain the project name (or None if no project is given). THe barcode will contain all Ns. The undetermined sample will not be aggregated with any other sample.

metrics.tsv

This file is always produced and contains a small number of summary statistics across the demultiplexing run:

Column	Description
control_reads_omitted	The number of reads that were omitted for being control reads.
failing_reads_omitted	The number of reads that were omitted for having failed QC.
total_templates	The total number of template reads that were output.

most_frequent_unmatched.tsv

This file is optional and will only be produced if --most-unmatched-to-output is not set to zero. It contains the (approximate) counts of the most prevelant observed barcode sequences that did not match to one of the expected barcodes.

Column	Description
barcode	The observed barcode sequence.
count	The approximate number of times that barcode sequences was observed.

sample_barcode_hop_metrics.tsv

This file is only output for dual-indexed runs. It contains frequently observed barcodes that are unexpected combinations of expected barcodes. For example if two samples are present with barcodes AA-CC and GG-TT, this file would report on observations of AA-TT and GG-CC if seen.

Column	Description
barcode	The observed barcode sequence.
count	The approximate number of times that barcode sequences was observed.

Advance Usage

Single Sample

It is possible to run sgdemux on a single sample without demultiplexing, in order to make use of the remaining functionality such as filtering control reads, extracting UMIs, etc. This mode is enabled by providing a sample metadata file that contains a single sample, with no barcode sequence. For example:

Sample_ID,Sample_Barcode
lone_sample,

The Sample_Barcode column must still be present, but empty for the sample. When running in this mode:

• All reads are assigned to the single sample
• No Undetermined files are created
• Sample barcodes, if read, will be inserted into the headers of the output FASTQ reads