Rapid haploid variant calling and core genome alignment

Torsten Seemann

Snippy finds SNPs between a haploid reference genome and your NGS sequence reads. It will find both substitutions (snps) and insertions/deletions (indels). It will use as many CPUs as you can give it on a single computer (tested to 64 cores). It is designed with speed in mind, and produces a consistent set of output files in a single folder. It can then take a set of Snippy results using the same reference and generate a core SNP alignment (and ultimately a phylogenomic tree).

% snippy --cpus 16 --outdir mysnps --ref Listeria.gbk --R1 FDA_R1.fastq.gz --R2 FDA_R2.fastq.gz
<cut>
Walltime used: 3 min, 42 sec
Results folder: mysnps
Done.

% ls mysnps
snps.vcf snps.bed snps.gff snps.csv snps.tab snps.html 
snps.bam snps.txt reference/ ...

% head -5 mysnps/snps.tab
CHROM  POS     TYPE    REF   ALT    EVIDENCE        FTYPE STRAND NT_POS AA_POS LOCUS_TAG GENE PRODUCT EFFECT
chr      5958  snp     A     G      G:44 A:0        CDS          41/600 13/200 ECO_0001  dnaA replication protein DnaA missense_variant c.548A>C p.Lys183Thr
chr     35524  snp     G     T      T:73 G:1 C:1    tRNA  -   
chr     45722  ins     ATT   ATTT   ATTT:43 ATT:1   CDS   -                    ECO_0045  gyrA DNA gyrase
chr    100541  del     CAAA  CAA    CAA:38 CAAA:1   CDS                        ECO_0179      hypothetical protein
plas      619  complex GATC  AATA   GATC:28 AATA:0  
plas     3221  mnp     GA    CT     CT:39 CT:0      CDS                        ECO_p012  rep  hypothetical protein

% snippy-core --prefix core mysnps1 mysnps2 mysnps3 mysnps4 
Loaded 4 SNP tables.
Found 2814 core SNPs from 96615 SNPs.

% ls core.*
core.aln core.tab core.tab core.txt core.vcf

Install Conda or Miniconda:

conda install -c conda-forge -c bioconda snippy

Install HomeBrew (Mac OS X) or LinuxBrew (Linux).

brew install brewsci/bio/snippy

This will install the latest version direct from Github. You'll need to add Prokka's bin directory to your $PATH.

cd $HOME
git clone https://github.com/tseemann/snippy.git
$HOME/snippy/bin/snippy --help

Ensure you have the desired version:

snippy --version

Check that all dependencies are installed and working:

snippy --check

• a reference genome in FASTA or GENBANK format (can be in multiple contigs)
• sequence read file(s) in FASTQ or FASTA format (can be .gz compressed) format
• a folder to put the results in

⚠️ ❌ Snippy 4.x does NOT produce the following files that Snippy 3.x did

If you supply a Genbank file as the --reference rather than a FASTA file, Snippy will fill in these extra columns by using the genome annotation to tell you which feature was affected by the variant:

The variant calling is done by Freebayes.
The key parameters under user control are:

• --mincov - the minimum number of reads covering a site to be considered (default=10)
• --minfrac - the minimum proportion of those reads which must differ from the reference
• --minqual - the minimum VCF variant call "quality" (default=100)

If you run Snippy with the --report option it will automatically run snippy-vcf_report and generate a snps.report.txt which has a section like this for each SNP in snps.vcf:

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>LBB_contig000001:10332 snp A=>T DP=7 Q=66.3052 [7]

         10301     10311     10321     10331     10341     10351     10361
tcttctccgagaagggaatataatttaaaaaaattcttaaataattcccttccctcccgttataaaaattcttcgcttat
........................................T.......................................
,,,,,,  ,,,,,,,,,,,,,,,,,,,,,t,,,,,,,,,,t,,t,,,,,,,,,,,,,,,,g,,,,,,,g,,,,,,,,,t,
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, .......T..................A............A.......
.........................A........A.....T...........    .........C..............
.....A.....................C..C........CT.................TA.............
,a,,,,,a,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,t,t,,,g,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
,,,,,ga,,,,,,,c,,,,,,,t,,,,,,,,,,g,,,,,,t,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
                            ............T.C..............G...............G......
                                                    ,,,,,,,g,,,,,,,,g,,,,,,,,,,,
                                                           g,,,,,,,,,,,,,,,,,,,,

If you wish to generate this report after you have run Snippy, you can run it directly:

cd snippydir
snippy-vcf_report --cpus 8 --auto > snps.report.txt

If you want a HTML version for viewing in a web browser, use the --html option:

cd snippydir
snippy-vcf_report --html --cpus 16 --auto > snps.report.html

It works by running samtools tview for each variant, which can be very slow if you have 1000s of variants. Using --cpus as high as possible is recommended.

• --rgid will set the Read Group (RG) ID (ID) and Sample (SM) in the BAM and VCF file. If not supplied, it will will use the --outdir folder name for both ID and SM.

• --mapqual is the minimum mapping quality to accept in variant calling. BWA MEM using 60 to mean a read is "uniquely mapped".

• --basequal is minimum quality a nucleotide needs to be used in variant calling. We use 13 which corresponds to error probability of ~5%. It is a traditional SAMtools value.

• --maxsoft is how many bases of an alignment to allow to be soft-clipped before discarding the alignment. This is to encourage global over local alignment, and is passed to the samclip tool.

• --mincov and --minfrac are used to apply hard thresholds to the variant calling beyond the existing statistical measure.. The optimal values depend on your sequencing depth and contamination rate. Values of 10 and 0.9 are commonly used.

• --targets takes a BED file and only calls variants in those regions. Not normally needed unless you are only interested in variants in specific locii (eg. AMR genes) but are still performing WGS rather than amplicon sequencing.

• --contigs allows you to call SNPs from contigs rather than reads. It shreds the contigs into synthetic reads, as to put the calls on even footing with other read samples in a multi-sample analysis.

If you call SNPs for multiple isolates from the same reference, you can produce an alignment of "core SNPs" which can be used to build a high-resolution phylogeny (ignoring possible recombination). A "core site" is a genomic position that is present in all the samples. A core site can have the same nucleotide in every sample ("monomorphic") or some samples can be different ("polymorphic" or "variant"). If we ignore the complications of "ins", "del" variant types, and just use variant sites, these are the "core SNP genome".

• a set of Snippy folders which used the same --ref sequence.

To simplify running a set of isolate sequences (reads or contigs) against the same reference, you can use the snippy-multi script. This script requires a tab separated input file as follows, and can handle paired-end reads, single-end reads, and assembled contigs.

# input.tab = ID R1 [R2]
Isolate1	/path/to/R1.fq.gz	/path/to/R2.fq.gz
Isolate1b	/path/to/R1.fastq.gz	/path/to/R2.fastq.gz
Isolate1c	/path/to/R1.fa		/path/to/R2.fa
# single end reads supported too
Isolate2	/path/to/SE.fq.gz
Isolate2b	/path/to/iontorrent.fastq
# or already assembled contigs if you don't have reads
Isolate3	/path/to/contigs.fa
Isolate3b	/path/to/reference.fna.gz

Then one would run this to generate the output script. The first parameter should be the input.tab file. The remaining parameters should be any remaining shared snippy parameters. The ID will be used for each isolate's --outdir.

% snippy-multi input.tab --ref Reference.gbk --cpus 16 > runme.sh

% less runme.sh   # check the script makes sense

% sh ./runme.sh   # leave it running over lunch

It will also run snippy-core at the end to generate the core genome SNP alignment files core.*.

The core.full.aln file is a FASTA formatted mutliple sequence alignment file. It has one sequence for the reference, and one for each sample participating in the core genome calculation. Each sequence has the same length as the reference sequence.

You can remove all the "weird" characters and replace them with N using the included snippy-clean_full_aln. This is useful when you need to pass it to a tree-building or recombination-removal tool:

% snippy-clean_full_aln core.full.aln > clean.full.aln
% run_gubbins.py -p gubbins clean.full.aln
% snp-sites -c gubbins.filtered_polymorphic_sites.fasta > clean.core.aln
% FastTree -gtr -nt clean.core.aln > clean.core.tree

• If you want to mask certain regions of the genome, you can provide a BED file with the --mask parameter. Any SNPs in those regions will be excluded. This is common for genomes like M.tuberculosis where pesky repetitive PE/PPE/PGRS genes cause false positives, or masking phage regions.
• If you use the snippy --cleanup option the reference files will be deleted. This means snippy-core can not "auto-find" the reference. In this case you simply use snippy-core --reference REF to provide the reference in FASTA format.

Sometimes you will have far more sequencing depth that you need to call SNPs. A common problem is a whole MiSeq flowcell for a single bacterial isolate, where 25 million reads results in genome depth as high as 2000x. This makes Snippy far slower than it needs to be, as most SNPs will be recovered with 50-100x depth. If you know you have 10 times as much data as you need, Snippy can randomly sub-sample your FASTQ data:

# have 1000x depth, only need 100x so sample at 10%
snippy --subsample 0.1  ...
<snip>
Sub-sampling reads at rate 0.1
<snip>

If you are looking for specific SNPs, say AMR releated ones in particular genes in your reference genome, you can save much time by only calling variants there. Just put the regions of interest into a BED file:

snippy --targets sites.bed ...

Sometimes one of your samples is only available as contigs, without corresponding FASTQ reads. You can still use these contigs with Snippy to find variants against a reference. It does this by shredding the contigs into 250 bp single-end reads at 2 × --mincov uniform coverage.

To use this feature, instead of providing --R1 and --R2 you use the --ctgs option with the contigs file:

% ls
ref.gbk mutant.fasta

% snippy --outdir mut1 --ref ref.gbk --ctgs mut1.fasta
Shredding mut1.fasta into pseudo-reads.
Identified 257 variants.

% snippy --outdir mut2 --ref ref.gbk --ctgs mut2.fasta
Shredding mut2.fasta into pseudo-reads.
Identified 413 variants.

% snippy-core mut1 mut2 
Found 129 core SNPs from 541 variant sites.

% ls
core.aln core.full.aln ...

This output folder is completely compatible with snippy-core so you can mix FASTQ and contig based snippy output folders to produce alignments.

The de novo assembly process attempts to reconstruct the reads into the original DNA sequences they were derived from. These reconstructed sequences are called contigs or scaffolds. For various reasons, small errors can be introduced into the assembled contigs which are not supported by the original reads used in the assembly process.

A common strategy is to align the reads back to the contigs to check for discrepancies. These errors appear as variants (SNPs and indels). If we can reverse these variants than we can "correct" the contigs to match the evidence provided by the original reads. Obviously this strategy can go wrong if one is not careful about how the read alignment is performed and which variants are accepted.

Snippy is able to help with this contig correction process. In fact, it produces a snps.consensus.fa FASTA file which is the ref.fa input file provided but with the discovered variants in snps.vcf applied!

However, Snippy is not perfect and sometimes finds questionable variants. Typically you would make a copy of snps.vcf (let's call it corrections.vcf) and remove those lines corresponding to variants we don't trust. For example, when correcting Roche 454 and PacBio SMRT contigs, we primarily expect to find homopolymer errors and hence expect to see ins more than snp type variants.

In this case you need to run the correcting process manually using these steps:

% cd snippy-outdir
% cp snps.vcf corrections.vcf
% $EDITOR corrections.vcf
% bgzip -c corrections.vcf > corrections.vcf.gz
% tabix -p vcf corrections.vcf.gz
% vcf-consensus corrections.vcf.gz < ref.fa > corrected.fa

You may wish to iterate this process by using corrected.fa as a new --ref for a repeated run of Snippy. Sometimes correcting one error allows BWA to align things it couldn't before, and new errors are uncovered.

Snippy may not be the best way to correct assemblies - you should consider dedicated tools such as PILON or iCorn2, or adjust the Quiver parameters (for Pacbio data).

Sometimes you are interested in the reads which did not align to the reference genome. These reads represent DNA that was novel to your sample which is potentially interesting. A standard strategy is to de novo assemble the unmapped reads to discover these novel DNA elements, which often comprise mobile genetic elements such as plasmids.

By default, Snippy does not keep the unmapped reads, not even in the BAM file. If you wish to keep them, use the --unmapped option and the unaligned reads will be saved to a compressed FASTQ file:

% snippy --outdir out --unmapped ....

% ls out/
snps.unmapped.fastq.gz ....

The name Snippy is a combination of SNP (pronounced "snip") , snappy (meaning "quick") and Skippy the Bush Kangaroo (to represent its Australian origin)

Snippy is free software, released under the GPL (version 2).

Please submit suggestions and bug reports to the Issue Tracker

• Perl >= 5.12
• Perl Modules: Time::Piece (core with modern Perl), Bioperl >= 1.6
• bwa mem >= 0.7.12
• minimap2 >= 2.0
• samtools >= 1.7
• bedtools >= 2.0
• bcftools >= 1.7
• GNU parallel >= 2013xxxx
• freebayes >= 1.1 (freebayes, freebayes-parallel, fasta_generate_regions.py)
• vcflib >= 1.0 (vcfstreamsort, vcfuniq, vcffirstheader)
• vt >= 0.5
• readseq >= 2.0
• snpEff >= 4.3
• samclip >= 0.2
• seqtk >= 1.2
• snp-sites >= 2.0
• wgsim >= 1.8 (for testing only - wgsim command)

For Linux (compiled on Ubuntu 16.04 LTS) and macOS (compiled on High Sierra Brew) some of the binaries, JARs and scripts are included.