DELLY is an integrated structural variant prediction method that can detect deletions, tandem duplications, inversions and translocations at single-nucleotide resolution in short-read massively parallel sequencing data. It uses paired-ends and split-reads to sensitively and accurately delineate genomic rearrangements throughout the genome.

• BamTools (https://github.com/pezmaster31/bamtools)
• Boost C++ Libraries (www.boost.org)
• OpenMP (www.openmp.org)
• zlib compression library (www.zlib.net)
• kseq library to parse FASTA/FASTQ (http://lh3lh3.users.sourceforge.net/parsefastq.shtml)

The easiest way to get DELLY is to download the source code from the DELLY github release page. Then you need to install Boost, BamTools and KSEQ and update the paths in the Makefile. Building DELLY just requires

make -B src/delly

Alternatively, a statically linked binary for Linux 64-bit is available here.

DELLY supports parallel computing using the OpenMP API (www.openmp.org).

make PARALLEL=1 -B src/delly

There is also a statically linked, multi-threaded binary for Linux 64-bit available under releases.

You can set the number of threads using the environment variable OMP_NUM_THREADS in your shell.

export OMP_NUM_THREADS=10

DELLY primarily parallelizes on the sample level. Hence, OMP_NUM_THREADS should be always smaller or equal to the number of input samples. As a rule of thumb you should consider an additional memory demand of about 1G per thread for human samples, about 100MB per thread for drosophila and less than 50MB per thread for yeast.

DELLY just needs one bam file for every sample and the reference genome to identify split-reads. The output is in vcf format. The SV type can be DEL, DUP, INV or JMP for deletions, tandem duplications, inversions and translocations, respectively.

./src/delly -t DEL -o del.vcf -g <ref.fa> <sample1.sort.bam> ... <sampleN.sort.bam>

Each bam file is assumed to be one sample. If you do have multiple bam files for a single sample please merge these bams using tools such as Picard and tag each library with a ReadGroup. To save runtime it is advisable to exclude telomere and centromere regions. For human, DELLY ships with such an exclude list.

./src/delly -t DEL -x human.hg19.excl.tsv -o del.vcf -g <ref.fa> <sample1.sort.bam> ... <sampleN.sort.bam>

If you omit the reference sequence DELLY skips the split-read analysis. The vcf file follows the vcf specification and all output fields are explained in the vcf header.

grep "^#" del.vcf

Delly ships with a small python script to annotate differential read-depth between SV carrier and non-carrier samples. This script is primarily meant as an example of how you can filter and annotate the final DELLY vcf file further.

python diffRC.py -t DEL -v del.vcf -o del.rd.vcf

There are also external packages that consume VCF files with per-sample genotype likelihoods. The arfer package annotates, for instance, Hardy-Weinberg Equilibrium and the inbreeding coefficient, which is useful for selecting polymorphic sites in the genome across a population.

cat del.rd.vcf | ./arfer/arfer > del.rd.arfer.vcf

• What about translocations?
  Not yet supported in DELLY v0.1.2. Please use the old single sample DELLY version available here.

• What is the smallest SV size DELLY can call?
  This depends on the sharpness of the insert size distribution. For an insert size of 200-300bp with a 20-30bp standard deviation, DELLY starts to call reliable SVs >=300bp.

• Can DELLY be used on a non-diploid genome?
  Yes and no. The SV site discovery works for any ploidy. However, the genotyping follows the classical hom. reference, het. and hom. alternative scheme.

• Is there any visualization of the called SVs?
  No, DELLY does not produce any graphical output. However, there are many viewers such as IGV that do visualize discordantly mapped paired-ends.

• How can DELLY be used to call somatic SVs?
  Run DELLY jointly on the cancer data and the matched control sequencing data. Ideally, you include many control samples in a single run because assuming that any reference mapping artifact is recurrent, multiple control samples from different patients will help you to catch these reference-biases more easily. In the end, one just filters the tumor SVs against all SVs present in any of the control genomes. For copy-number variable events (CNVs), such as deletions and tandem duplications the additional normalized read-count genotype field (RC) can help to differentiate complex rearrangements that do not necessarily show a read-depth change from simple CNVs that have a supporting read-depth signal. Do not run multiple tumor genomes together since overlapping somatic SVs might have different coordinates in different tumor genomes. The setup should be tumor.bam + control.bam(s).

• Can DELLY be used on bwa mem alignments?
  DELLY can be used with bwa mem but you have to mark shorter split-read alignments as secondary alignments using the '-M' option, e.g. bwa mem -M ref.fa file.fq.gz.

• Are non-unique alignments, multi-mappings and/or multiple split-read alignments allowed?
  DELLY expects two alignment records in the bam file for every paired-end, one for the first and one for the second read. Multiple split-read alignment records of a given read are allowed if and only if one of them (e.g. the longest split alignment) is a primary alignment whereas all others are marked as secondary (flag 0x0100).

Tobias Rausch, Thomas Zichner, Andreas Schlattl, Adrian M. Stuetz, Vladimir Benes, Jan O. Korbel.
DELLY: structural variant discovery by integrated paired-end and split-read analysis.
Bioinformatics 2012 28: i333-i339.