We’re excited to introduce our Best Practices recommendations for calling variants on RNAseq data. These recommendations are based on our classic DNA-focused Best Practices, with some key differences in the early data processing steps, as well as in the calling step.

Best Practices workflow for RNAseq

This workflow is intended to be run per-sample; joint calling on RNAseq is not supported yet, though that is on our roadmap.

Please see the new document here for full details about how to run this workflow in practice.

In brief, the key modifications made to the DNAseq Best Practices focus on handling splice junctions correctly, which involves specific mapping and pre-processing procedures, as well as some new functionality in the HaplotypeCaller.

Now, before you try to run this on your data, there are a few important caveats that you need to keep in mind.

Please keep in mind that our DNA-focused Best Practices were developed over several years of thorough experimentation, and are continuously updated as new observations come to light and the analysis methods improve. We have only been working with RNAseq for a few months, so there are many aspects that we still need to examine in more detail before we can be fully confident that we are doing the best possible thing.

For one thing, these recommendations are based on high quality RNA-seq data (30 million 75bp paired-end reads produced on Illumina HiSeq). Other types of data might need slightly different processing. In addition, we have currently worked only on data from one tissue from one individual. Once we’ve had the opportunity to get more experience with different types (and larger amounts) of data, we will update these recommendations to be more comprehensive.

Finally, we know that the current recommended pipeline is producing both false positives (wrong variant calls) and false negatives (missed variants) errors. While some of those errors are inevitable in any pipeline, others are errors that we can and will address in future versions of the pipeline. A few examples of such errors are given in this article as well as our ideas for fixing them in the future.

We will be improving these recommendations progressively as we go, and we hope that the research community will help us by providing feedback of their experiences applying our recommendations to their data. We look forward to hearing your thoughts and observations!

Return to top


Are the authors of GATK BaseRecalibrator concerned about Post-transcriptional Modifications in RNA-seq? Reverse Transcriptases have difficulty correctly reading modified nucleotides. The Reverse Transcriptase may produce an error at a modified nucleotide when making the cDNA. Illumina will then read the resulting cDNA correctly (and give high quality score). Thus, even though the Illumina reads are correctly reporting the base in the cDNA (with high quality scores), it will be "wrong" compared to the reference, and not masked by dbSNP since it is only a Post-transcriptional modification. This will severely reduce the resulting empirical quality scores calculated by BaseRecalibrator. For DNA-seq, BaseRecalibrator masks "--knownSites" of polymorphism when calculating empirical Quality scores. The "--knownSites" is usually a VCF from e.g. dbSNP. In RNA-seq, do the authors of GATK recommend any kind of VCF with known Post-transcriptional modifications in mRNA?

Thu 6 Mar 2014


I'm afraid we don't have any recommendations for this -- in our hands BQSR performed normally on RNAseq data, but we haven't tested for this specifically. The size of potential effect is linked to how random vs. not random the errors might be, and what is the rate at which they occur. The more random and the lower the rate, the less noticeable any potential effect.

Thu 6 Mar 2014


Hello, Does expression level effects how many variants be found?

Thu 6 Mar 2014


@sbombin Potentially, yes. If you have different expression of two alleles, and one is very low, you may not be able to call them correctly -- e.g. you may call the wrong genotype, or miss the call altogether (if it looks hom-ref). Also, anything that is not expressed at all will typically not be callable.

Thu 6 Mar 2014


Hi Geraldine and others, I'm trying to run this RNA-seq variant calling pipeline on human tumor data (we have quite a few), and it seems like variant calling using Haplotypecaller is our rate limiting step. While alignment and preprocessing are quick, Haplotypecaller runs a bit slower. Is there any way to speed up the process? Could we potentially exclude indels (not of interest at the moment) to speed up the runs?

Thu 6 Mar 2014


@dnousome HaplotypeCaller runs slower since it is a haplotype based calling method. HCaller is loosing time when doing denovo assembly of haplotypes in the active region, to find the most likely haplotype. Based on the number of samples and size of the genome and the computing power you have, it will take considerable time. On my desktop with 4 cores, 12 gb RAM, 205 mb reference genome and 12 samples, HaplotypeCaller takes about 2-3 days. If you use UnifiedGenotyper it will work way faster, like couple of hours for the same parameters-sample combination, I discussed earlier. I am not sure how much indel exclusion helps with the run of the speed. But, see this: http://gatkforums.broadinstitute.org/wdl/discussion/1975/how-can-i-use-parallelism-to-make-gatk-tools-run-faster HCaller only support -nct but isn't quite stable. But, on the positive note: I am used server with large RAM capacity (several gbs) and in this case the HCaller perfoms way much faster. In the report files I see that quite a high amount of memory was allocated to HCaller. I am not sure why the use of large RAM isn't quite discussed to speed up HCaller.

Thu 6 Mar 2014


Hey all, it's not possible to exclude indels from HaplotypeCaller, and we don't recommend using UnifiedGenotyper because when you don't model for indels, you will make errors in SNPs that are near indels. We recommend parallelizing execution of HC using data scatter gather.

Thu 6 Mar 2014


Dear Geraldine How does this pipeline is accurate? Would you show me any paper or data, comparing whole exome/genome sequence and this pipeline? Thanks

Thu 6 Mar 2014


@satoshi, we have not published any benchmarking or comparisons on this pipeline ourselves. Some of our collaborators have published a paper that I think may be related, which may include comparative results. We've linked to it previously; you may find it and related papers by looking for Ami Levy-Moonshine, Brian Haas and Timothy Tickle as authors.

Thu 6 Mar 2014

At a glance

Follow us on Twitter

GATK Dev Team


Slides, tutorial materials from #GATK workshop in Leuven, Belgium posted at https://t.co/LaXzONIp1C
19 Feb 17
RT @EMBL_ABR: Register your interest now in training in @gatk_dev @BroadGenomics https://t.co/gwEzuwOZYP to be run in Aus by @BioplatformsA…
19 Feb 17
#GATK Support team appreciation day -- say hi and get a prize! https://t.co/J71I4FL9zA
23 Jan 17
RT @EricTopol: Great explainer of human genome structural variation https://t.co/BaAWWmi30A @broadinstitute https://t.co/282chNv4oI
21 Jan 17
@broadinstitute @TwinTwoG Sure, happy to help on the GATK forum; just post your question at https://t.co/6z2bZ2AeN0
10 Jan 17

Our favorite tweets from others

@froggleston @dgmacarthur Sounds like ExAC is reaching Uber stage. ‘Uber but for pizza’. ‘ExAC but for wheat’.
14 Jan 17
#ESRenpeinture grad school - postdoc - after postdoc https://t.co/o3vQMgBDgk
6 Jan 17
Really happy to have you again this year! @VIBLifeSciences https://t.co/8rg5VQ2fbX
3 Jan 17
Currently in a time-out for saying that duck fat had a certain "je ne sais quack" at the thanksgiving dinner table.
25 Nov 16
@dgmacarthur @BioMickWatson @StevenNHart @splon There's even a shop near Broad that apparently fixes Hail code erro… https://t.co/IZ4BcgRZYE
19 Nov 16
See more of our favorite tweets...
Search blog by tag

appistry ashg ashg16 benchmarks best-practices bug bug-fixed cancer cloud cluster cnv collaboration commandline community compute conferences cram cromwell depthofcoverage diagnosetargets error forum gatk3 gatk4 genotype-refinement genotypegvcfs google grch38 gvcf haploid haplotypecaller hg38 holiday hts htsjdk ibm intel java8 job job-offer jobs license meetings mutect mutect2 ngs outreach pairhmm parallelism patch performance picard pipeline plans ploidy polyploid poster presentations printreads profile promote release release-notes rnaseq runtime saas script sequencing service slides snow speed status support syntax talks team terminology topstory troll tutorial unifiedgenotyper vcf-gz version-highlights versions vqsr wdl webinar workflow workshop xhmm