ikmb/deepvariant repository overview

⁠IKMB DeepVariant Pipeline

This pipeline performs an end-to-end variant calling, starting from raw fastQ files to a multi-sample VCF. It has been pre-configured for the CCGA MedCluster.

⁠Running the pipeline

This pipeline requires nextflow and singularity, so make sure these modules are loaded.

To run the pipeline, a typical command will looks as follows:

nextflow run ikmb/deepvariant --samples Samples.csv --genome GRCh38

These options are further explained in the following:

⁠Options

⁠--samples

This option expects a sample sheet in CSV format that specifies information on the subject, sample, details of the library and the location of the paired-end files.

A script is included that takes a folder full of fastQ files and creates a compliant sample sheet. This script requires ruby and can be downloaded here⁠

ruby samplesheet_from_folder.rb --folder /path/to/reads > Samples.csv

Please note that the script makes some assumptions about the naming structure of the FastQ files (typically Illumina-generated data should work - but please check manually to see if the information makes sense!).

⁠--joint_calling

Perform joint calling of all samples in this pipeline run (otherwise, only per-sampe VCFs are produced without considering variant sites in other samples.

⁠--genome

The name of the genome assembly version to use. Allowed options are:

• GRCh37 (1000Genomes reference with decoys)
• GRCh38 (Current human genome WITHOUT alt contigs, as recommended by Heng Li)
• hg38 (Current human genome with all alt contigs, as distributed by the BROAD)

Please note that all assemblies come with pre-defined calling regions, as provided by the BROAD institute. These were manually curated to exclude regions that cannot be reliably called with short read data

⁠Accuracy

Deepvariant has been shown to be highly accurate to the Genome-in-a-Bottle benchmark sets. Using in-house produced WGS data from GIAB sample NA12878, the following concordance scores were achieved within the GIAB high confidence intervals:

type TP FP FN RECALL PRECISION

SNP 3040675 1048 2825 0.999 1.0

INDEL 479765 19271 20038 0.96 0.961