STRkit - short tandem repeat genotyping with long reads

STRkit is a short tandem repeat (STR) genotyping and analysis toolkit for long read sequencing data, especially
PacBio HiFi data. The STRkit software package is written in Python and is available in the PyPI package registry or as
a Docker container.

STRkit's main advantages over other callers include:

• Better performance in some benchmarking assessments versus other tools;
• An on-the-fly SNV calling and local phasing step for unphased aligned reads;
• A license which allows use on any long-read sequencing data, versus TRGT whose
  license restricts it to only PacBio data.

If you use STRkit in published work, please cite our paper in Genome Research:

Read-level genotyping of short tandem repeats using long reads and single-nucleotide variation with STRkit.
David R Lougheed, Tomi Pastinen, Guillaume Bourque. Genome Res.
DOI: 10.1101/gr.280766.125

The code for the software itself is also available and citeable in Zenodo: https://doi.org/10.5281/zenodo.12689906

Officially-supported platforms: macOS (aarch64), Linux (amd64/aarch64)

Note: Currently, it is unlikely that STRkit will work on "plain" Windows due to upstream dependencies.
To run STRkit on Windows, please use Windows Subsystem for Linux or
use our Docker image (see Installation as a Docker container).

Table of Contents

• Installation
  • Via PyPI
  • As a Docker container
• Commands
  • strkit call: Genotype caller with bootstrapped confidence intervals
    • Features
    • Usage
    • Further documentation on the STRkit caller, including output format
  • strkit visualize: Call visualizer
  • strkit mi: Mendelian inheritance analysis
    • Usage
    • Further documentation
  • strkit convert: STR catalog conversion
    • Usage
• Copyright and License
  • Notice
  • Exceptions

Installation

Via PyPI

STRkit requires Python 3.10+ and can be installed from PyPI via pip
with the following command:

python -m pip install strkit

You may need to install the Rust toolchain (version 1.88.0 or newer)
and a C compiler (e.g., gcc, clang), as well as cmake, to compile the strkit_rust_ext wheel,
although prebuilt wheels for this module are available for some platforms. Compiling the wheel may take quite
a long time (in the tens of minutes).

On Digital Research Alliance of Canada/Compute Canada clusters, this involves loading a few modules:

module load rust/1.91.0 clang/18.1.8 python/3.11 scipy-stack/2026a parasail/2.6.2
python -m pip install strkit

STRkit should then be available in your Python environment as a command-line tool:

strkit --help

As a Docker container

STRkit is also available as a Docker container, stored
in the GitHub Container Registry.

It can be pulled using the following command:

docker pull ghcr.io/davidlougheed/strkit:latest

Then, STRkit commands can be run mostly as normal using the Docker image:

docker run -it ghcr.io/davidlougheed/strkit --help

For example, to run the strkit call subcommand in Docker, including writing file output (via binding input files and
mounting a read/writeable folder for outputs):

docker run \
  -v inputs:/inputs:ro \
  -v out:/out \
  -it ghcr.io/davidlougheed/strkit call \
  /inputs/file.bam \
  --loci /inputs/loci.bed --ref /inputs/ref.fa --incorporate-snvs /inputs/dbsnp.vcf.gz \
  --vcf /out/calls.vcf --no-tsv

Inside the inputs directory, STRkit would (given these parameters) expect the following files to be present:

• file.bam: A readset for the sample to be called.
• file.bam.bai: A BAI index for the readset.
• ref.fa: The reference genome for the reads in FASTA format.
• ref.fa.fai: A FAIDX-format index for the reference genome.
• dbsnp.vcf.gz: A VCF of SNVs to use as a catalogue for calling SNVs to aid in the STR genotyping process.
• dbsnp.vcf.gz.tbi: A Tabix-format index for the dbSNP VCF.
• loci.bed: A tab-delimited BED-format catalogue of loci (see the
  Caller catalog format & choosing a catalog document for more information).

For more information on call parameters, see the below section on strkit call and the
Advanced caller usage and configuration document.

Commands

strkit call: Genotype caller with bootstrapped confidence intervals

A Gaussian mixture model tandem repeat genotype caller for long read data.
STRkit is tuned specifically for high-fidelity long reads, although other
long read data should still work.

Features:

• Performant, vectorized (thanks to parasail)
  estimates of repeat counts from high-fidelity long reads and a supplied
  catalog of TR loci and motifs.
• Re-weighting of longer reads, to compensate for their lower likelihood of observation.
  • Whole-genome and targeted genotyping modes to adjust this re-weighting.
• Incorporation of single-nucleotide variation (SNVs) for better and faster calling plus
  additional downstream analysis possibilities.
  • Recommended for HiFi data and ONT R10 data only. In my testing, this worsens runtime and call quality for
    ONT ultra-long-read data, but speeds up the tool and improves call quality for HiFi/ONT R10 data.
  • Important note: This functionality is best for whole-genome STR surveying. If hunting for rare / low-coverage
    expansions, or using targeted sequencing data, it may be best to NOT USE an SNV catalog to avoid discording
    reads with short flanking regions.
• Parallelized for faster computing on clusters and for ad-hoc fast analysis of single samples.
• 95% confidence intervals on calls via a user-configurable optional parametric bootstrapping process.

Usage:

See all parameters and example usage with a Slurm cluster:
Advanced caller usage and configuration

EXAMPLE USAGE

# For the dbSNP VCF used below for SNV incorporation, see https://ftp.ncbi.nih.gov/snp/organisms/human_9606/VCF/
# (00-common_all.vcf.gz)
#
# "Accurate reads" here means HiFi / ONT R10 duplex reads, but in practice may also include ONT R10 simplex reads.

strkit call \
  path/to/read/file.bam \  # [REQUIRED] One indexed read file (BAM/CRAM)
  --realign \  # If using accurate reads, enable this to enable local realignment / read recovery. Good for detecting expansions, but slows down calling.
  --ref path/to/reference.fa.gz \  # [REQUIRED] Indexed FASTA-formatted reference genome
  --loci path/to/loci.bed \  # [REQUIRED] TRF-formatted (or 4-col, with motif as last column) sorted list of loci to genotype
  --annotation-file path/to/reference-annotations.gff.gz \  # Optional GFF3/GTF file to provide locus annotations (transcript/exon ID)
  --incorporate-snvs path/to/dbsnp/00-common_all.vcf.gz \   # If you want, specify a SNV catalogue to help phase STRs & speed up calling
  --vcf my-calls.vcf \  # Calculate consensus sequences for alleles and output a .vcf (or .vcf.gz) with call data
  --seed 183 \  # Fixed random number generator seed for replicability
  --processes 10 \  # Number of parallel processes to use; DEFAULT: 1
  --no-tsv  # If VCF output is enabled as above, we don't need TSV genotype output to stdout (which is the default)

REGARDING ALIGNMENTS

Ideally, you should be using a read file aligned with parameters tuned for tandem repeats.
PacBio provides a
recommended workflow
for CCS alignment in this scenario. However, regular aligned readsets are fine and have been tested
extensively.

WITH ACCURATE READS

If you're using accurate long reads (e.g., HiFi, ONT R10) as input, use the --realign option
to enable re-aligning soft-clipped reads to the reference genome. This will add a bit of
performance overhead, but can help recover some reads of interest (e.g., expansions).

In practice, this option may also aid calling with slightly-less-accurate reads, especially if
you are looking for expansions and do not mind slower performance if the reads do not align well
to the reference genome.

WITH LOW-QUALITY READS

If you're using low-quality long reads (e.g., older ONT data), some non-default parameters may help
get the best genotyping performance:

• Reads with tandem repeat + flanking region PHRED quality below 13 are filtered out by default.
  You may wish to lower this to avoid discarding reads by setting --min-avg-phred to a lower
  value like 5, or you can effectively disable this filter by setting it to 1.
• We have observed some cases of 'false positive' expansion-esque reads with ONT R9 data. The
  --force-gm-filter option can help filter these out, at the cost of reducing STRkit's
  sensitivity to true expansions.

WITH HAPLOTAGGED READS

If you want to incorporate haplotagging from an alignment file (HP tags) into the
process, which should speed up runtime and potentially improve calling results, you must pass
the --use-hp flag.

WITH LOW-COVEREAGE ALIGNMENT DATA

If you're calling STRs using low-coverage alignments, you may want to:

• lower the minimum number of reads per locus (--min-reads) to 2 (or even 1 if you're fine
  with some guaranteed allelic dropout), and
• lower the minimum number of reads per allele (--min-allele-reads) to 1 to avoid dropout as
  much as possible.
• potentially, if you desire more calls at the cost of quality, lower the minimum tandem repeat +
  flanking region PHRED quality required (from its default of 13) via the --min-avg-phred
  parameter.

Note: --min-reads must be greater than or equal to --min-allele-reads.

WITH TARGETED SEQUENCING DATA

If you're running STRkit with targeted sequencing data, you should put STRkit in targeted mode with the --targeted
flag, which slightly tweaks read re-sampling behaviour and increases the --max-reads default from 250 to 500.

LOOKING FOR LARGE EXPANSIONS

If you're running STRkit for the purpose of finding pathogenic STR expansions or expansions of
interest, the following parameters may help increase expansion sensitivity, at the cost of
computational performance or a potential bias towards heterozygous calls:

• Make sure to NOT enable --force-gm-filter, even if using lower-quality reads. Enabling this option can filter
  out artifact reads with false-positive expansion-like read counts, but it will also lower sensitivity for true
  expansions with low read support.
• Enable --realign, even with lower-quality reads. Reads containing expansions can be
  soft-clipped by aligners like minimap2; the --realign flag can help recover these reads.
• Tune --gm-filter-expansion-ratio to a slightly lower value, like 3.0. This will keep larger alleles with low read
  support if they are larger than --gm-filter-expansion-ratio times the size of the smaller allele.

REGARDING SNV INCORPORATION

If you want to incorporate SNV calling into the process, which speeds up runtime and gives
marginally better calling results, you must provide an indexed, bgzip-compressed SNV catalog
VCF which matches your reference genome. You can find dbSNP VCFs at
https://ftp.ncbi.nih.gov/snp/organisms/human_9606/VCF/.
The file for GRCh38 is called 00-common_all.vcf.gz as of time of writing.
Note that this does not need to be an SNV call file for your sample, specifically; just one
which has positions, reference/alternate alleles, and the ID field populated.

REGARDING ANNOTATION

If you would like to have the IDs of genome features (from a .gff3 or .gtf file) which
overlap STR loci in your strkit call output (JSON or VCF), you can pass a sorted, Tabix-indexed
annotation file path to strkit call with the --annotation-file <path> flag (replacing <path>
with the path to the annotation file.) This could be used for downstream analysis to find STR
variants that overlap, for example, coding sequences or gene UTRs.

For more information, see also documentation on the Output formats.

REGARDING OUTPUT

If you want to output a full call report, you can use the --json output-file.json argument to
specify a path to output a more detailed JSON document to. This document contains 99% CIs, peak
labels, and some other information that isn't included in the normal TSV file. If you want this
file to be indented and human-readable, use the --indent-json flag in addition to --json ....

If you want to output a VCF file (STRs and SNVs if called; currently not phased), use the
--vcf ... argument. If you pass --vcf stdout, the VCF will be written to stdout instead of a
file.

For more information, see also documentation on the Output formats.

Note on output locus coordinates

STRkit can adjust reference coordinates of STR loci to include regions it thinks should be part of
the definition. Thus, output coordinates may not match the original BED catalog definition. This is
done for two purposes:

• To include slightly mismapped indels that may lie outside the catalog-defined region
• To try to achieve a more consistent absolute copy number between the reference region and the
  sample data

Using the --respect-ref flag (see Caller usage), users may suppress
this behaviour.

REGARDING REFERENCE GENOMES

The reference genome provided must be BGZipped and indexed using samtools faidx:

# Starting from a .fa:
bgzip my-reference.fa  # Replaces .fa with a .fa.gz file
samtools faidx my-reference.fa.gz  # Generates a .fai index file

OTHER PARAMETERS

See the 'Caller catalog format & choosing a catalog' page for more on
how to format a locus catalog or choose from existing available catalogs.

Further documentation on the STRkit caller, including output format:

• Advanced caller usage and configuration
• Caller catalog format & choosing a catalog
• Output formats

strkit visualize: Call visualizer

STRkit bundles a call visualization tool which takes as input a BAM file and
a JSON call file from using the --json flag with strkit call.

It starts a web server on your local machine; the visualizations can be
interacted with in a web browser.

To use the tool, run the following command:

strkit visualize path/to/my-alignment.bam \ 
  --ref hg38 \  # or hg19
  --json path/to/my-calls.json \
  -i 1  # 1-indexed offset in JSON file for locus of interest. Default is 1 if left out.

This will output something like the following:

 * Serving Flask app 'strkit.viz.server' (lazy loading)
 * Environment: production
   WARNING: This is a development server. Do not use it in a production deployment.
   Use a production WSGI server instead.
 * Debug mode: on
 * Running on http://localhost:5011 (Press CTRL+C to quit)
...

You can then go to the URL listed, http://localhost:5011, on your local machine
to see the visualization tool:

STRkit browser histogram, showing an expansion in the HTT gene.

The same expansion, shown in the igv.js browser. Note the insertions on
the left-hand side in most reads, and the heterozygous copy number pattern.

To exit the tool, press Ctrl-C in your command line window as mentioned in
the start-up instructions.

Further documentation on the STRkit web visualization tool:

• Using strkit visualize remotely

strkit mi: Mendelian inheritance analysis

Using trio data, candidate de novo STR mutations (or genotyping errors/dropout rates) can be discovered
by looking at inheritance patterns. This tool provides a few different ways to do this, via:

• Mendelian inheritance % (MI) calculations for many common TR genotyping tools for both long/short reads,
  including support for genotyping methods which report confidence intervals.
• Reports of loci (potentially of interest) which do not respect MI.

Usage

For a basic JSON report on Mendelian inheritance with a trio of STRkit VCFs (compressed and indexed with BGZip), use
something like the following command:

# In addition to summary figures on Mendelian inheritance, this tool outputs loci which do not respect MI, which may be 
# useful as candidate de novo mutations. The --mismatch-out-mi flag controls which form of MI metric is used for 
# deciding which loci to output. Options for this flag are:
#   strict (strict copy number MI),
#   pm1 (copy number MI ± 1 repeat unit),
#   ci_95 (copy number 95% confidence interval),
#   ci_99 (copy number 99% confidence interval),
#   seq ([allele] sequence MI),
#   sl ([allele] sequence length MI),
#   sl_pm1 ([allele] sequence length MI ± 1 base pair)
strkit mi \
  --caller strkit-vcf \
  --json mi-report.json \
  --mismatch-out-mi seq \
  child-calls.vcf.gz \
  mother-calls.vcf.gz \
  father-calls.vcf.gz
# This will also output a TSV report to stdout. If this is not desired, use --no-tsv to suppress TSV output.

For other options and what they do, run strkit mi (with no other arguments) or strkit mi --help.

Further documentation

For more information on what kind of analyses can be done with this data, see the
Trio analyses with STRkit page.

strkit convert: STR catalog conversion

STRkit takes as input a four-or-more-column BED file, structured like one of the following:

contig  start end [0 or more extraneous columns] NNN
contig  start end [0 or more extraneous columns] ID=locus1;MOTIF=NNN

Any extraneous columns are removed, (internally) leaving a four-column STR locus representation.
Some other tools, e.g., Straglr, also take a four-column STR
BED as locus catalog input. However, other formats representing a catalog of STRs exist:

• Tandem Repeats Finder outputs a TSV/BED with a lot
  of information. This can be used as-is with STRkit, but it's safer for other tools to convert to
  a four-column BED format.
• TRGT uses a custom repeat definition format,
  which can specify more advanced STR structures.

Usage

The strkit convert sub-command requires an input format (trf or trgt), an output format
(many, see strkit convert --help), and an input file. Output is written to stdout.

Note: Not all input/output format pairs have available converter functions; an error will be
printed to stderr if one does not exist.

For example, to convert from a TRF BED to a TRGT repeat definition BED file:

strkit convert --in-format trf --out-format trgt in_file.trf.bed > out_file.bed

To convert from a TRGT repeat definition file to a STRkit BED, including locus IDs if set:

strkit convert --in-format trgt --out-format strkit in_file.trgt.bed > out_file.strkit.bed

Where possible, multiple motifs will be reconciled using IUPAC codes,
which STRkit supports.

Note that TRGT can represent STRs with complex structure that STRkit cannot. In these cases, the first motif in the TRGT
locus definition will be used and a message will be logged to stderr.

Copyright and License

• 2021-2023: © David Lougheed (DL) and McGill University 2021-2023 (versions up to and including 0.8.0a1),
  created during graduate research by DL.
• 2023+: (versions beyond 0.8.0a1):
  • Portions © DL and McGill University 2021-2023
  • Portions © McGill University 2024-2026
  • Portions © DL 2024-2026

Notice

This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program. If not, see https://www.gnu.org/licenses/.

Exceptions

Some exclusions to this license apply; specifically portions of
strkit/viz/templates/browser.html and
the STRkit logo files (./docs/images/strkit_logo_small.png
and ./strkit/viz/static/logo.png.)

The STRkit logo is © David Lougheed 2022, and was designed by Evelyn Lougheed. It is not licensed
under the terms of the GPL 3.0; it is instead licensed under the terms of the
CC BY-ND 4.0.

Portions of viz/templates/browser.html copyright (C) 2021-2022 Observable, Inc.
Used under the terms of the ISC license.