201. ReduceReads will no longer be in GATK 30

IMPORTANT: This is the legacy GATK documentation. This information is only valid until Dec 31st 2019. For latest documentation and forum click here

created by Geraldine_VdAuwera

on 2014-03-04

This is an important heads-up regarding the GATK 3.0 release.

The purpose of the ReduceReads algorithm was to enable joint analysis of large cohorts by the UnifiedGenotyper. The new workflow for joint discovery, which involves doing a single-sample pass with the HaplotypeCaller in gVCF mode followed by a joint analysis on multiple sample gVCFs, renders the compression step obsolete.

In addition, based on our most recent analyses, we have come to the conclusion that the quality of variant calls made on BAMs compressed with ReduceReads is inferior to the standard targeted by GATK tools. In comparison, the results obtained with the new workflow are far superior.

For these reasons, we have made the difficult decision to remove the ReduceReads tool from version 3.0 of the toolkit. To be clear, reduced BAMs will NOT be supported in GATK 3.0.

We realize that this may cause some disruption to your existing workflows, and for that we apologize. Please understand that we are driven to provide tools that produce the best possible results. Now that all the data is in, we have found that the best results cannot be achieved with reduced BAMs, so we feel that the best thing to do is to remove this inferior tool from the toolkit, and promote the new tools.

As always we welcome your comments, and we look forward to showing you how the new calling workflow will yield superior results.

Updated on 2014-03-04

From jpitt on 2014-03-04

Hi @Geraldine_VdAuwera,

Thanks for the heads up on this! Since I saw this post I’ve looked around without much success. Is HaplotypeCaller’s gVCF and subsequent multi-sample gVCF calling implemented in the current 2.8.1 build of GATK, or do we need to wait until 3.0 is released?

Also, for multi-sample calling you say that ReducedReads + UnifiedGenotyper perform subotimally compared to HaplotypeCaller + gVCF. Does this mean that the calls obtained from ReducedReads bams are inherently poor? More specifically, are ReducedReads + UnifiedGenotyper calls of much lower quality than regular bams + UnifiedGenotyper calls? Assuming that HaplotypeCaller gVCF is coming in version 3.0, would your team advise against using ReducedReads bams in the meantime?

Thanks!

From Geraldine_VdAuwera on 2014-03-04

Hi @jpitt,

The gVCF (or reference model) is already implemented in 2.8, though it is further improved in the 3.0 code. However the tools for making joint analyses based on gVCFs of multiple samples (which is what replaces RR+UG) are not yet available; they will be in the new release. Which, if all goes well, will be released tomorrow! So I would advise waiting a day or two for the new version to be available.

> are ReducedReads + UnifiedGenotyper calls of much lower quality than regular bams + UnifiedGenotyper calls?

It’s all relative, but let’s say that they are inferior enough (particularly to the results we get with the new method) that we absolutely encourage you to switch over to the new method.

From BretH on 2014-03-05

How will you be filtering out low-information reads (like pcr duplicates)? Or are these included in the joint varient analysis step?

From Geraldine_VdAuwera on 2014-03-05

Hi @BretH,

We mark duplicates using Picard tools in pre-processing, same as previously. This causes GATK tools to ignore dupes in subsequent analyses.

From BretH on 2014-03-06

Thank you for your answer.