V'DJer uses customized read extraction, assembly and V(D)J rearrangement detection and filtering to produce contigs representing the most abundant portions of the BCR repetoire. V'DJer operates upon bulk short read mRNA-Seq data allowing sequence analysis typically done using specialized long read sequencing. The software can be run on BCR heavy (IgH) and light (Igk/IgL) chains.

V'DJer accepts a sorted and indexed BAM file mapped to hg38 as input. We currently use STAR for alignment. At the end of a run assembled contigs will appear in vdj_contigs.fa and reads aligned to those contigs are written to stdout

The V'DJer output is suitable for use by downstream quantification tools such as RSEM.

Use a recent release: https://github.com/mozack/vdjer/releases

To compile, just cd into the vdjer directory and type make. An executable named vdjer will be created.

To date, V'DJer has been tested only on linux systems.

Pre-built human indices and references can be downloaded from here: https://github.com/mozack/vdjer/releases/download/v0.10_reference/vdjer_human_references.tar.gz

The archive must be untarred and decompressed for use by V'DJer

A reference directory is included for each chain type.

V'DJer operates on paired end reads. Single end reads are not currently supported. When mapping with STAR, unmapped reads must be included in the BAM file (this is not STAR's default behavior). Include the unmapped reads using the following param:

--outSAMunmapped Within

Be sure to sort and index the BAM file.

The following runs vdjer on the input star.sort.bam file with vdj_contigs.fa generated in the current working directory and read alignments written to vdjer.sam:

vdjer --in star.sort.bam --t 8 --ins 175 --chain IGH --ref-dir vdjer_human_references/igh > vdjer.sam 2> vdjer.log

The above runs on the IgH chain with read length of 50, 8 threads and median insert length of 175.

In cases where a sample has low BCR expression levels, one may opt to run V'DJer using more sensitive settings. Running sensitive mode on samples with abundant BCR expression levels may result in extremely high computational costs. Example command:

vdjer --in star.sort.bam --t 8 --ins 175 --chain IGH --ref-dir vdjer_human_references/igh --k 25 
--mq 60 --mf 2 --rs 25 --ms 2 --mcs -5.5 > vdjer.sam 2> vdjer.log 

Decreasing mq and mf from the defaults results in less aggressive graph pruning. Reducing rs and ms results in less aggressive coverage based filtering Decreasing mcs allows for more exhaustive graph traversal

The values used in this example match those used when running sensitive mode in the V'DJer paper.

See demo.bash and quant_demo.bash under the demo directory for an example of running V'DJer.

https://doi.org/10.1093/bioinformatics/btw526