Assemble bacterial isolate genomes from Illumina paired-end reads

The SPAdes genome assembler has become the de facto standard de novo genome assembler for Illumina whole genome sequencing data of bacteria and other small microbes. SPAdes was a major improvement over previous assemblers like Velvet, but some of its components can be slow and it traditionally did not handle overlapping paired-end reads well.

Shovill is a pipeline which uses SPAdes at its core, but alters the steps before and after the primary assembly step to get similar results in less time. Shovill also supports other assemblers like SKESA, Velvet and Megahit, so you can take advantage of the pre- and post-processing the Shovill provides with those too.

⚠️ Shovill is for isolate data only, primarily small haploid organisms. It will NOT work on metagenomes or larger genomes. Please use Megahit directly instead.

1. Estimate genome size and read length from reads (unless --gsize provided)
2. Reduce FASTQ files to a sensible depth (default --depth 100)
3. Trim adapters from reads (with --trim only)
4. Conservatively correct sequencing errors in reads
5. Pre-overlap ("stitch") paired-end reads
6. Assemble with SPAdes/SKESA/Megahit with modified kmer range and PE + long SE reads
7. Correct minor assembly errors by mapping reads back to contigs
8. Remove contigs that are too short, too low coverage, or pure homopolymers
9. Produce final FASTA with nicer names and parseable annotations

% shovill --outdir out --R1 test/R1.fq.gz --R2 test/R2.fq.gz

<snip>
Final assembly in: test/contigs.fa
It contains 17 (min=150) contigs totalling 169611 bp.
Done.

% ls out

contigs.fa   contigs.gfa   shovill.corrections  
shovill.log  spades.fasta

% head -n 4 out/contigs.fa

>contig00001 len=52653 cov=32.7 corr=1 origname=NODE_3 date=20180327 sw=shovill/1.0.1
ATAACGCCCTGCTGGCCCAGGTCATTTTATCCAATCTGGACCTCTCGGCTCGCTTTGAAGAAT
GAGCGAATTCGCCGTTCAGTCCGCTGGACTTCGGACTTAAAGCCGCCTAAAACTGCACGAACC
ATTGTTCTGAGGGCCTCACTGGATTTTAACATCCTGCTAACGTCAGTTTCCAACGTCCTGTCG

brew install brewsci/bio/shovill
shovill --check

Using Homebrew will install all the dependencies for you: Linux or MacOS

conda install -c conda-forge -c bioconda -c defaults shovill
shovill --check

Using Bioconda will install all the dependencies for you on MacOS and Linux.

The Docker recipe is generously maintained by Curtis Kapsak and the StaPH-B workgroup.

# Docker
docker pull staphb/shovill:latest
docker run staphb/shovill:latest shovill --help

# Singularity
singularity build shovill.sif docker:https://staphb/shovill:latest
singularity exec shovill.sif shovill --help

git clone https://github.com/tseemann/shovill.git
./shovill/bin/shovill --help
./shovill/bin/shovill --check

You will need to install all the dependencies manually:

• SPAdes >= 3.11 (prefer >= 3.14)
• SKESA
• MEGAHIT
• Velvet >= 1.2
• Lighter
• FLASh
• SAMtools >= 1.3 (prefer >= 1.10)
• BWA MEM
• KMC
• seqtk
• pigz. Pigz should be available with your OS distribution.
• Pilon (Java).
• Trimmomatic (Java)
• samclip

Note that you will need to make pilon and trimmomatic executables. You can make a simple wrapper for each that just passes the shell arguments.

This is most important output file - the final, corrected assembly. It contains entries like this:

>contig00001 len=263154 cov=8.9 corr=1 origname=NODE_1 date=20180327 sw=shovill/0.9
>contig00041 len=339 cov=8.8 corr=0 origname=NODE_41 date=20180327 sw=shovill/0.9

The sequence IDs are named as per the --namefmt option, and the comment field is a series of space-separated name=value pairs with the following meanings:

SYNOPSIS
  De novo assembly pipeline for Illumina paired reads
USAGE
  shovill [options] --outdir DIR --R1 R1.fq.gz --R2 R2.fq.gz
GENERAL
  --help          This help
  --version       Print version and exit
  --check         Check dependencies are installed
INPUT
  --R1 XXX        Read 1 FASTQ (default: '')
  --R2 XXX        Read 2 FASTQ (default: '')
  --depth N       Sub-sample --R1/--R2 to this depth. Disable with --depth 0 (default: 150)
  --gsize XXX     Estimated genome size eg. 3.2M <blank=AUTODETECT> (default: '')
OUTPUT
  --outdir XXX    Output folder (default: '')
  --force         Force overwite of existing output folder (default: OFF)
  --minlen N      Minimum contig length <0=AUTO> (default: 0)
  --mincov n.nn   Minimum contig coverage <0=AUTO> (default: 2)
  --namefmt XXX   Format of contig FASTA IDs in 'printf' style (default: 'contig%05d')
  --keepfiles     Keep intermediate files (default: OFF)
RESOURCES
  --tmpdir XXX    Fast temporary directory (default: '/tmp/tseemann')
  --cpus N        Number of CPUs to use (0=ALL) (default: 8)
  --ram n.nn      Try to keep RAM usage below this many GB (default: 16)
ASSEMBLER
  --assembler XXX Assembler: skesa velvet megahit spades (default: 'spades')
  --opts XXX      Extra assembler options in quotes eg. spades: "--untrusted-contigs locus.fna" ... (default: '')
  --kmers XXX     K-mers to use <blank=AUTO> (default: '')
MODULES
  --trim          Enable adaptor trimming (default: OFF)
  --noreadcorr    Disable read error correction (default: OFF)
  --nostitch      Disable read stitching (default: OFF)
  --nocorr        Disable post-assembly correction (default: OFF)

Giving an assembler too much data is a bad thing. There comes a point where you are no longer adding new information (as the genome is a fixed size), and only adding more noise (sequencing errors). Most assemblers seem to be happy with ~150x depth, so Shovill will downsample your FASTQ files to this depth. It estimates depth by dividing read yield by genome size.

The genome size is needed to estimate depth and for the read error correction stage. If you don't provide --gsize, it will be estimated via k-mer frequencies using mash. It doesn't need to be a perfect estimate, just in the right ballpark.

This will keep all the intermediate files in --outdir so you can explore and debug.

By default it will attempt to use all available CPU cores.

Shovill will do its best to keep memory usage below this value, but it is not guaranteed. If you are on a HPC cluster, you should make sure you tell your job submission engine a value higher than this.

By default it will use SPAdes, but you can also choose Megahit or SKESA. These are much faster than SPAdes, but give lesser assemblies. If you use SKESA you can probably use --noreadcorr and --nocoor because it has some of that functionality inbuilt and is conservative.

If you want to provide some assembler-specific parameters you can use the --opts parameter. Make sure you quote the parameters so they get passed as a single string eg. For --assembler spades you might use --opts "--sc --untrusted-contigs similar_genome.fasta" or --opts '--sc'.

A series of kmers are chosen based on the read length distribution. You can override this with this option.

These env-vars will be used as defaults instead of the built-in defaults. You can use the normal command line option to override them still.

• Does shovill accept single-end reads?

  No, but it might one day.

• Do you support long reads from Pacbio or Nanopore?

  No, this is strictly for Illumina paired-end reads only. Try use Flye. CANU, or Redbean.

• Why does Shovill crash?

  Shovill has a lot of dependencies. If any dependencies are not installed correctly it will die. Spades also doesn't handle --cpus > 16 very well - try giving more RAM.

• Can I assemble metagenomes with Shovill?

  No. Please use dedicated tools like Minia 3.x or Megahit. Shovill uses the estimated genome size for many dynamic settings related to read error correction, read subsampling etc.

Please file questions, bugs or ideas to the Issue Tracker

GPLv3

Not published yet.

• Torsten Seemann
• Web: https://tseemann.github.io/
• Twitter: @torstenseemann
• Blog: The Genome Factory

• Jason Kwong
• Simon Gladman
• Anders Goncalves da Silva