SCAN-SNV

Somatic genotyper for SNV discovery in whole genome amplified single cells.

Note

We now recommend users of SCAN-SNV to instead use our new tool SCAN2. SCAN2 contains all of the functionality of SCAN-SNV, but is faster and easier to use. SCAN2's VAF-based SNV calls (i.e., calls prior to mutational signature-based rescue with scan2 rescue) are produced using nearly the same procedure as SCAN-SNV. SCAN-SNV users should be aware that SCAN2 uses a more stringent calling cutoff by default (--target-fdr=0.01 in SCAN2 vs. 0.1 in SCAN-SNV).

What does SCAN-SNV do?

SCAN-SNV identifies somatic single nucleotide variants (sSNVs) from whole genome amplified single cell DNA-seq. SCAN-SNV was designed with multiple displacement amplification (MDA) in mind, but the principle should apply to any amplification method that can produce uneven amplification between homologous alleles.

All sSNVs identified by SCAN-SNV are assumed to be heterozygous since there is a very low likelihood of the same somatic mutation occurring twice at the same location and with the same base change. Other mechanisms of generating non-heterozygous somatic mutations exist (i.e., loss of heterozygosity), but these are not modelled by SCAN-SNV.

Does SCAN-SNV genotype germline SNVs?

No, SCAN-SNV does not call germline SNVs. The SCAN-SNV methodology will refuse to identify any SNV for which any mutation supporting reads are found in the (required) matched bulk sample. If deeply sequenced (i.e., ~30X or greater) bulk data is used, virtually all germline mutations will have at least one supporting read and thus will not be called.

The SCAN-SNV allele balance model can provide useful information for germline SNV discovery, however we currently do not plan to develop this further.

Installation

SCAN-SNV is distributed as a conda package. Installation requires the conda package management tool and a Linux-flavored OS.

Operating systems tested

• GNU/Linux, kernel version 3.10.0, CentOS 7. Note that pre-compiled SHAPEIT2 binaries are only made available for Linux systems, although in principle other phasing algorithms can be used instead.
• Ubuntu 16.04.4 LTS AWS instance.

NOTE This installation process and demo have been successfully run on an Amazon Web Services Ubuntu 16.04.4 LTS instance.

Installing miniconda

$ wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
$ bash Miniconda3-latest-Linux-x86_64.sh
# Accept the license by typing "yes"
# Choose an install prefix (the default is often fine)
# Choose to run conda init (enter yes a second time during script)
# Log-out and back in to source .bashrc and put conda on $PATH

Installing SCAN-SNV

Create a conda environment for SCAN-SNV

$ conda deactivate   # The "base" environment will be active after login
$ conda create -n scansnv
$ conda activate scansnv

Install the scansnv package

$ conda install -c bioconda -c conda-forge/label/cf201901 -c jluquette scansnv

Register your GATK installation

$ wget 'https://software.broadinstitute.org/gatk/download/auth?package=GATK-archive&version=3.8-1-0-gf15c1c3ef' -O GenomeAnalysisTK-3.8-1-0-gf15c1c3ef.tar.bz2
$ tar xjvf GenomeAnalysisTK-3.8-1-0-gf15c1c3ef.tar.bz2
$ gatk-register GenomeAnalysisTK-3.8-1-0-gf15c1c3ef/GenomeAnalysisTK.jar
# Test the install
$ gatk --version
# Above should print 3.8-1-0-gf15c1c3ef

Downloading external data dependencies

SCAN-SNV has been tested on the NCBI human reference build 37.

Download reference genome.

$ wget ftp://[email protected]/bundle/b37/human_g1k_v37_decoy.fasta.gz
$ wget ftp://[email protected]/bundle/b37/human_g1k_v37_decoy.fasta.fai.gz
$ wget ftp://[email protected]/bundle/b37/human_g1k_v37_decoy.dict.gz

Download dbSNP. Note that dbSNP build 147 (common variants only) was used in the publication. However, NCBI does not guarantee long term hosting of dbSNP builds, so we recommend downloading the version of dbSNP included in the Broad's GATK resource bundle. To use other builds of dbSNP, you will need to generate a tribble index (see below).

$ wget ftp://[email protected]/bundle/b37/dbsnp_138.b37.vcf.gz
$ wget ftp://[email protected]/bundle/b37/dbsnp_138.b37.vcf.idx.gz

Download SHAPEIT's haplotype reference panel.

$ wget https://mathgen.stats.ox.ac.uk/impute/1000GP_Phase3.tgz
$ wget https://mathgen.stats.ox.ac.uk/impute/1000GP_Phase3_chrX.tgz

Unzip everything and move the chrX SHAPEIT files into the main SHAPEIT directory.

$ gunzip *.gz
$ tar xzvf 1000GP_Phase3.tgz
$ tar xzvf 1000GP_Phase3_chrX.tgz
$ mv genetic_map_chrX_* 1000GP_Phase3_chrX* 1000GP_Phase3

Running the SCAN-SNV demo

Download the demo chr22 BAMs.

$ wget http://compbio.med.harvard.edu/scan-snv/hunamp.chr22.bam
$ wget http://compbio.med.harvard.edu/scan-snv/hunamp.chr22.bam.bai
$ wget http://compbio.med.harvard.edu/scan-snv/il-12.chr22.bam
$ wget http://compbio.med.harvard.edu/scan-snv/il-12.chr22.bam.bai

Run SCAN-SNV. Replace instances of /path/to/... with the paths downloaded above. This demo runs in about 5 minutes on a single core machine by restricting analysis to a 1 MB segment of chr22 and by using an impractically coarse grid for covariance function fitting.

scansnv \
    --ref /path/to/human_g1k_v37_decoy.fasta \
    --dbsnp /path/to/dbsnp_138.b37.vcf \
    --shapeit-panel /path/to/1000GP_Phase3 \
    --regions 22:30000001-31000000 \
    --output-dir demo \
    --bam hunamp hunamp.chr22.bam \
    --bam h25 il-12.chr22.bam  \
    --sc-sample h25 \
    --bulk-sample hunamp \
    --abmodel-chunks 1 \
    --abmodel-samples-per-chunk 10000 \
    --abmodel-hsnp-chunk-size 50 \
    --hsnp-spikein-replicates 5 \
    --joblimit 1 --resume

See scansnv -h for more details on arguments.

After SCAN-SNV completes, single sample results are available in the Rdata file demo/scansnv/[single_cell_sample_name]/somatic_genotypes.rda. SNVs that pass SCAN-SNV's calling thresholds will have pass=TRUE in the somatic data frame (see below).

NOTE: a VCF output option is forthcoming.

# Called sSNVs can be extracted from the data frame via
R> load('demo/scansnv/[single_cell_sample_name]/somatic_genotypes.rda')
R> somatic[somatic$pass,]
# The demo should not produce any passing variants.

WARNING!

• Sample names associated with BAMs MUST match the SM tag in the BAM.
• The conda environment (named scansnv in these instructiosn) must always be conda activated before running SCAN-SNV.
• Real world analyses will require parallelization.
  • On a machine with multiple cores, increasing the --joblimit parameter will run multiple parts of the analysis in parallel.
  • For clusters with distributed resource management software (e.g., SLURM), SCAN-SNV exposes Snakemake's parallelization options --cluster and --drmaa.

Using a custom dbSNP version

Generating a Tribble index for dbSNP

dbSNP VCFs must be indexed by Tribble (not tabix) for GATK. The dbSNP found in the GATK's resource bundle is already indexed. If you wish to use a different dbSNP version, the file can be indexed by igvtools.

$ conda install -c bioconda igvtools
$ igvtools index /path/to/your/dbsnp.vcf