Test-drive the GATK tools and Best Practices pipelines on Terra
Check out this blog post to learn how you can get started with GATK and try out the pipelines in preconfigured workspaces (with a user-friendly interface!) without having to install anything.
IMPORTANT -- Bug alert for GATK4 GenomicsDBImport
We have identified a major bug in the GenomicsDBImport tool that affects all released beta versions of GATK4 up to 4.beta.5 (inclusive). The bug occurs under specific conditions (detailed below) and causes the output of joint calling to be scrambled across samples, i.e. the sample names will not be associated with the correct sample data. For example, the data for
sample1 may be labeled as
sample3. The good news is that the results are recoverable as long as you have a record of the exact parameters used in the original command.
So if you have used this tool, please read the detailed description of the bug conditions and recovery procedure below. We apologize for any inconvenience this may cause you.
Going forward, everyone who plans to use GenomicsDBImport should upgrade to GATK4 version 4.beta.6 or later.
Conditions under which the bug occurs
The bug occurs when all of the following conditions are met:
- You used a version of GenomicsDBImport from before GATK4 version 4.beta.6 (for nightlies, up to 4.beta.5-66-g9609cb3-SNAPSHOT);
- You used the
--batchSizeargument with a setting other than 0;
- You imported multiple batches, meaning you had more input GVCFs than what your batch size was set to;
- The input GVCFs were not sorted according to Java’s natural sorting of strings (see details below).
That last point applies if you specified files using the
-V argument or if you used a
sampleNameMap file. Note that Java sorts strings lexicographically, not in a numerically aware fashion, so
sample1, sample2, …, sample9, sample10, sample11 would be considered "Out of order".
Given the following
HG00096 HG00096.g.vcf.gz NA19625 NA19625.g.vcf.gz HG00268 HG00268.g.vcf.gz
This command using
./gatk-launch GenomicsDBImport --sampleNameMap samples.txt --batchSize 2 --genomicsDBWorkspace workspace -L chr21
or this one using
./gatk-launch GenomicsDBImport -V HG00096.g.vcf.gz -V NA19625.g.vcf.gz -V HG00268.g.vcf.gz --batchSize 2 --genomicsDBWorkspace workspace -L chr1
would be affected by the bug and produce corrupted output.
We have a new tool in GATK4 version 4.beta.6 called FixCallSetSampleOrdering. This tool will take a corrupted callset and reassign the correct names. It requires that you use the EXACT SAME settings you used in your GenomicsDBImport command for the
--batchSize argument and for the method and order you used to specify input files. For the examples shown above, the recovery command would be:
./gatk-launch FixCallSetSampleOrder --sampleNameMap samples.txt --batchSize 2 -O fixedCallSet.vcf
To be clear, if you specified the input GVCFs with
-V originally, you’ll need to create a sample name map with the ordering you used in your command in order to run the fixup tool.
Note that the settings aren’t recorded anywhere in the final output, so if you didn't keep a record of what those settings were, you may need to re-call your dataset from the original BAMs. However, please contact us before doing so. We are working on a new fingerprinting tool that should be able to recover the correct names in many cases without recalling. It will also be able to validate that the sample names are correct.
We apologize again for any disruption this may cause to your work.