Currently I'm performing whole analysis (pipeline from *.fastq to *.vcf) of 41 samples (targeted NGS). I rely on GATK best practices, however with some modifications. I decided to use the following tools:

bwa mem (with mem - alternate haplotypes)

RevertSam (Picard)
MergeBamAlignment (Picard)
MarkDuplicates (Picard)
BaseRecalibrator (GATK)
AnalyzeCovariates (GATK)
SortSam (Picard)

#call, filter and annotate variants
HaplotypeCaller (GATK)
GenotypeGVCFs (GATK)
snpEff, SnpSift

Nevertheless, after marking the duplicates with MarkDuplicates about 99% of reads have been marked as duplicated, hence the coverage decreased from several hundreds to literally up to five. So I decided to omit this step.

Have you ever encountered a similar problem? What is your opinion about this?

Best Regards, Adam


The Agilent Haloplex custom target enrichment kit (HaloPlex Target Enrichment for Illumina Kit) has been used to prepare the libraries, sequencing done using Illumina HiSeq 2500. Targeted NGS of all exons (only) of 90 genes.

Here I provide how the final bam look like. This is the result without duplicate marking, otherwise there're only several reads per exon... enter image description here

  • $\begingroup$ can you clarify what you mean by targeted NGS? What are you targeting/capturing and how were the sequencing libraries prepared? $\endgroup$
    – conchoecia
    Feb 14, 2019 at 6:32
  • 3
    $\begingroup$ I believe, using haloplex target capture, you expect reads to have identical begin and end coordinates, so imho in this case you should skip duplicate marking/removal. $\endgroup$ Feb 14, 2019 at 7:43
  • 1
    $\begingroup$ Wouter is right, Haloplex is PCR based, consequently you expect high duplication. Do not deduplicate this data prior to analysis. $\endgroup$
    – user192
    Feb 14, 2019 at 12:06
  • $\begingroup$ Thanks for help! By the way, what do you think about GATK as a tool for such analysis. For me it's nice but a bit too complicated. $\endgroup$
    – Adamm
    Feb 14, 2019 at 12:39
  • $\begingroup$ Indeed after MarkDuplicates the coverage was up to 4. It was hard to distinguish between actual alternations and mismatched bases. $\endgroup$
    – Adamm
    Feb 15, 2019 at 5:55

2 Answers 2


Most duplicate marking programs do it by looking at the start and end location of hits. In this case, if two reads have identical start and end locations for both reads, then they will be considered [PCR] duplicates of each other. When running through a duplicate filter, you will only get one read for each unique pair.

This is much more likely for targeted sequencing, because the length covered by the template is much smaller. It's also more likely that you actually do have lots of PCR duplicates for the same reason.


If you're doing target capture, especially of a small region, it is entirely possible that there are only five unique reads. Target capture is molecularly finicky and sometimes the entire region is not captured.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.