Rapid generation of genetic imputation scripts for grid/cluster/local environments

molgenis-impute is a tool for rapid generation and submission of scripts that perform genetic imputation. The generated scripts contain all rigorous quality control checks, data chunking, error handling and reporting. The tool is designed to be simple and straight-forward, for researchers that want to do imputation now, without limiting the available options of imputation tools.

molgenis-impute performs three main actions:

• liftover (from hg18 to hg19 genomic reference)
• phasing of a study panel
• imputation

molgenis-impute runs in any 64-bit x86 Linux distribution and it requires the following tools:

• wget (or curl)
• tar
• unzip
• bunzip2
• g++
• java 1.6 or higher
• python 2.7
• numpy

For example, to set up the environment in Ubuntu you can run:

sudo apt-get update
sudo apt-get install -y git openjdk-8-jre g++ python-numpy unzip make zlib1g-dev 

Also note that imputation reference panels can take tens of GBs of disk space

To install MOLGENIS-impute download the latest release from here.

For the .tar.gz file, uncompress it with: tar zxvf < FILENAME > . For the .zip file, uncompress it with: unzip < FILENAME > .

To get the cutting edge latest version, clone this repository:

git clone https://github.com/kantale/molgenis-imputation.git 

Initially, run the following to download all necessary tools for imputation:

python molgenis-impute.py --dl_tools

By default the files will be installed in the molgenis_imputation directory in the current location. To change this use the --installation_dir < DIR > option.

The next step is to download a reference panel of your choice. To list all available reference panels either for direct use or for downloading, run:

python molgenis-impute.py --list

To download a reference panel, run: (you can skip this step if you only want to use the pre-installed testing reference panel named "test_reference")

python molgenis-impute.py --dl_reference < NAME >

where < NAME > is the name of the reference panel as provided by the --list option.

this will install the reference panel to the default location: molgenis_imputation/resources/imputationReference. To select a custom location use the --reference_dir < DIR > option.

Example:

python molgenis-impute.py --dl_reference GIANT.metabo.phase1_release_v3.20101123

ATTENTION This command will fail if you haven't installed the tools and datasets first, with the --dl_tools option.

A study panel should be in plink's PED and MAP format: http://pngu.mgh.harvard.edu/~purcell/plink/data.shtml . Moreover, the PED/MAP files should be splitted per chromosome and stored in a single directory. For example:

• my_study/chr1.ped
• my_study/chr1.map
• my_study/chr2.ped
• my_stydy/chr2.map
• ...

In this release we include a testing study panel in the directory: molgenis_imputation/resources/GWAS/small . For information about how this study was prepared check the molgenis_imputation/resources/GWAS/small/README.txt file.

In molgenis-impute, in order to use a directory with a study panel in any of the analysis use the option: --study and provide the absolute path of that directory.

Liftovering is the process of changing the genomic assembly of a dataset from one version (usual older) to another (newer). To liftover a study panel from the hg18 genome assembly to hg19, run the following:

python molgenis-impute.py --study < STUDY DIRECTORY > --output < OUTPUT DIRECTORY >  --action liftover

For Example:

python molgenis-impute.py --study `pwd`/molgenis_imputation/resources/GWAS/small/ --output `pwd`/results_liftover --action liftover

Under the hood molgenis-impute uses the liftOver tool from UCSC. The output will be stored in the directory defined with the --output option in ped/map format. The filenames will be:

• chr1.ped , chr1.map
• chr2.ped , chr2.map
• ...

The result of this process is in binary plink format. By default the liftover that is performed is from hg18ToHg19. You can change the assembly by using the argument --chain_file . The accepted values are: hg18tohg19 and hg18ToHg38. Alternatively you can specify your own chain filename with the same argument. Repositories of chain files for liftovering starting from hg18 and hg19 builds can be found here: http://hgdownload.cse.ucsc.edu/goldenPath/hg18/liftOver/ , http://hgdownload.cse.ucsc.edu/goldenPath/hg19/liftOver/ . For example you can download the hg18ToEquCab1.over.chain.gz , save it to a local directory and use the option: --chain_file /path/to/hg18ToEquCab1.over.chain.gz

Phasing is the process of determining the haplotype structure of genotype data. To phase a dataset it should be either in plink text format (PED/MAP) or binary (BED/BIM/FAM). The format is automatically detected. The command is:

python molgenis-impute.py --study < STUDY DIRECTORY > --output < OUTPUT DIRECTORY >  --action phase

For example:

python molgenis-impute.py --study `pwd`/results_liftover --output `pwd`/results_phase --action phase

Under the hood molgenis-impute uses the SHAPEIT tool. The output will be stored in the directory defined in the --output option in genotype/sample format.

To impute a phased dataset run the following command:

python molgenis-impute.py --study < PHASED STUDY DIRECTORY > --output < OUTPUT DIRECTORY >  --action impute --reference < REFERENCE NAME >

For example:

python molgenis-impute.py --study `pwd`/results_phase --reference test_reference --output `pwd`/results_impute --action impute

The options that this command takes are:

• < STUDY DIRECTORY > is the directory where the study panel exists. The study panel should be phased (preferrably with SHAPEIT) in the genotype/sample format.
• < OUTPUT DIRECTORY> is the directory where the output will be stored
• < REFERENCE NAME > is the name of the reference panel that will be used for the imputation. To get a list of all reference panels available run:

python molgenis-impute.py --list

Under the hood molgenis-impute uses Genotype Harmonizer for quality control and impute2 tool for imputation. This tool removes SNPs from the study that strand correction cannot be applied (for example an A/T SNP in the study that exists as A/C in the reference panel). It also generates a log file of all the performed checks that includes all removed markers. This file is saved in the defined output directory as: chrXYZ.log (XYZ is the number of chromosome, for example: chr1.log)

The imputation task is split in many chunks. The split is 2-dimensional: according to genomic position and according to samples:

• The genomic position split is per 5.000.000 distance. You can change this with the --position_batch_size option.
• The sample split is done so that each chunk should have approximately the same number of samples. The default setting is that each sample chunk should have at least 500 samples but not more than twice this value (1000=2*500). To change the default value of 500, use the --sample_batch_sizeoption. The location of the BASH script that splits the data is: tools/molgenis-pipelines-master/compute5/Imputation_impute2/protocols/impute2Imputation.sh (denoted with the comment: #START OF SAMPLE SPLITTING).

By default molgenis-impute will perform imputation for all chromosomes located in the reference panel. You can limit the imputation chromosomes with the option --chromosomes < comma separated values of chromosomes > For example: --chromosomes 1,3,8

If the reference panel is not in the default directory (the < current directory >/resources/imputationReference). Define the custom directory with the --reference_dir parameter. For example the following options: --reference_dir /my/custom/dir --reference 1000GP will assume that the reference panel is installed in /my/custom/dir/1000GP directory.

By default molgenis-impute assumes that java is in the PATH of the execution system. If this is not the case, use the option --java_executable to define the path to java executable. For example: --java_executable /path/to/java

The Phase and Impute Steps (Step 2 and 3) can be combined with the option --action phase_impute

python molgenis-impute.py --study < STUDY DIRECTORY > --reference < REFERENCE NAME > --output < OUTPUT DIRECTORY >  --action phase_impute  

For example:

python molgenis-impute.py --study `pwd`/results_liftover --reference test_reference --output `pwd`/results_impute --action phase_impute

To combine all three steps use the option: --action liftover_phase_impute

python molgenis-impute.py --study < STUDY DIRECTORY > --reference < REFERENCE NAME > --output < OUTPUT DIRECTORY > --action liftover_phase_impute

For example:

python molgenis-impute.py --study `pwd`/molgenis_imputation/resources/GWAS/small/ --reference test_reference --output `pwd`/results_impute --action liftover_phase_impute

The molgenis-impute distribution includes an example study panel. This panel is part of the HapMap3 release 2 dataset (first 100 samples, first 10Mbp) and is located in the resources/GWAS/small directory. For more info about this test dataset you can take a look at resources/GWAS/small/README.md. You can impute this dataset with a subset of GIANT release of 1000 Genomes Project that is also included in the distribution in the directory resources/imputationReference/test_reference/

• liftover from hg18 to hg19:

python molgenis-impute.py --study `pwd`/molgenis_imputation/resources/GWAS/small/ --output `pwd`/results_liftover --action liftover

• phase:

python molgenis-impute.py --study `pwd`/results_liftover --output `pwd`/results_phase --action phase

• impute:

python molgenis-impute.py --study `pwd`/results_phase --reference test_reference --output `pwd`/results_impute --action impute

The final results of this proccess will be at the results_impute directory. The `pwd` part in the paths is to make sure that the paths are absolute (pwd is the Linux command to Print the Working Directory).

To add a new reference panel create a new directory in molgenis_imputation/resources/imputationReference (or in any custom location in case you are using the --reference_dir option). The name of the directory will be the name of the new reference panel. In this directory, store the reference panel in Variant Called Format (VCF). The files should have .vcf extension. Moreover each chromosome should be in a separate file and the name of the file should have at any point a chr< CHROMOSOME NUMBER > part. The naming should be consistent for all files. For example:

• 1000GP_chr1.vcf
• 1000GP_chr2.vcf
• ...

You don't need to do anything else. The next time you run molgenis-impute.py it will detect the new files and do the appropriate conversions. Plase take note that some conversion take a considerable amount of time, specially for large vcf files.

This is the recommended way for installing a new reference panel. Alternatively, if you want to install your own .haps and .legend files, you can place them in a new directory under molgenis_imputation/resources/imputationReference. Each chromosome should be in a separate pair of files. If the files are uncompressed the extension should be .haps and .legend . Optionally, the files can be compressed with gzip and the files' extensions should be: .haps.gz and .legend.gz . For example: 1000_GP_chr1.haps.gz and 1000_GP_chr1.legend.gz. Finally either the .vcf or the compressed .vcf.gz should also exist in this directory for each chromosome. IMPORTANT: The .vcf.gz files should not be compressed with gzip, but with bgzip instead. bgzip is installed in tools/tabix-0.2.6/ .

There are more alternative ways to install a reference panel. In general molgenis-impute supports the following reference panels formats:

• SHAPEIT2 Format: Files should be per chromosome. Example: chr1_SHAPEIT.haps and chr1_SHAPEIT.sample (The _SHAPEIT part is necessary in order to make a distiction from impute2 .haps files)
• Impute2 Format: Files shoud be per chromosome. Example: chr1.haps , chr1.legend . There should also be a SINGLE .sample file.
• Impute2 Format compressed: Files should be per chromosome. Example: chr1.haps.gz , chr1.legend.gz . There should also be a SINGLE .sample file (not compressed). This is the default format of the reference panels available from impute2. For example: http://mathgen.stats.ox.ac.uk/impute/data_download_1000G_phase1_integrated_SHAPEIT2_9-12-13.html . Take note that some panels have the filename pattern of chr1.hap.gz instead of chr1.haps.gz . In this case you will need to rename the files.
• VCF: A single vcf file.
• VCF compressed. A single compressed (with bgzip NOT gzip) vcf file. Example panel.vcf.gz

For all these files molgenis-compute will try to do its best to apply the appropriate conversions before performing imputation. If you have a special request or you notice a bug please let me know!

To make use of the new reference in the imputation step use the option --reference < REFERENCE_NAME >. The < REFERENCE_NAME > is the name of the created directory under molgenis_imputation/resources/imputationReference . (If unsure run python molgenis-impute.py --list for a list of available reference panels)

• --installation_dir: set the installation directory for imputation tools and resources. Default: < currrent working dir >/molgenis_imputation
• --reference_dir: set the installation directory for the imputation reference panels. Default: < currrent working dir >/molgenis_imputation/resources/imputationReference
• --nosubmit: Do not submit for execution the generated scripts.
• --results: Same as --output
• --additional_shapeit_parameters: Additional parameters to pass to SHAPEIT2 tool. These parameters should be quoted with single(') or double (") quotation marks. For example: --additional_shapeit_parameters "--exclude-snp gwas.subset.site"
• --additional_impute2_parameters: Additional parameters to be passed to IMPUTE2 tool. These parameters should be quoted with single(') or double (") quotation marks. For example: --additional_impute2_parameters "-Ne 20000"

The reason why the values of the last two parameters shoud be in quotation marks is that otherwise these values would be mistaken as parameters of the molgenis-impute.py script.

After a successful submission, MOLGENIS-impute prints a random generated ID and the location of the generated scripts. For example:

RANDOM ID FOR THIS RUN WAS:  f0935de6
Generated scripts are saved in:  /home/ubuntu/molgenis-imputation/molgenis_imputation/generated/impute_f0935de6

This means that the generated temporary (or intermediate) files of the analysis are stored in the folder: < OUTPUT FOLDER >/tmp_< ID > . Where < OUTPUT FOLDER > is the folder declared with the --output parameter and < ID > is the generated ID . Inspection of these files can sometimes give important insights regarding the performed analysis. The location of the generated scripts contains all scripts that have been submitted to the cluster. The same location is used to save the standard output and standard error from the execution of the scripts (if the cluster supports this functionality). This location also has a submit.sh script that you can run in order to re-submit an analysis.

All scripts detect if the output files are in place and in case they are, the execution is skipped. This helps in cases when an execution get abruptly stopped, to resume from the last succesful execution step. By selecting a different results directory or deleting the generated results you can repeat the analysis.

This software is under the Simplified BSD Licese.

• Main development:
  • Alexandros Kanterakis
• Contributors:
  • Patrick Deelen
  • Freerk Van Dijk
  • George Byelas
  • Martijn Dijkstra
• Supervision:
  • Morris Swertz
• The code is also hosted in pypedia (http://www.pypedia.com/index.php/Imputation). To obtain the code of imputation.py from pypedia run:

curl http://www.pypedia.com/index.php?get_code=%23Imputation\(\)  > imputation.py 

Alexandros Kanterakis
[email protected]
Genetica
UMCG
Postbus 30 001
9700 RB Groningen
The Netherlands