From Genome Analysis Wiki
Jump to navigationJump to search
The printable version is no longer supported and may have rendering errors. Please update your browser bookmarks and please use the default browser print function instead.

RAREMETALWORKER is a tool for generating summary statistics for rare variants and gene level meta analyses using RAREMETAL.

If you feel this program is useful, please tell us your name and contact in this registration.

If you have any questions, please contact Shuang Feng sfengsph at umich dot edu or Goncalo Abecasis goncalo at umich dot edu.

Useful Wiki Pages

There are several pages in this Wiki that may be useful to RAREMETALWORKER users. Here are links to key pages:

Key Features

RAREMETALWORKER has the following features:

  • Takes genotypes from either PED file or VCF file.
  • Generates summary statistics for both related and unrelated individuals.
  • Generates linkage disequilibrium matrices summarizing covariance between single marker statistics using an adjustable sliding window.
  • Optionally handles related individuals using a kinship matrix derived from either pedigree or genotype data.
  • Has the option of fitting shared environment.
  • Can handle variants on Chromosome X.
  • Calculates QC statistics such as hwe pvalue, call rate and genomic control.
  • Automatically generate QQ and manhattan plots.

Software Download and Installation

Where to Download

We have tested our source code package on several platforms, including Linux/Ubuntu, Linux/Debian, Linux/RedHat, Linux/CentOS, Mac OS X 10.8, CygWIN64 and MinGW for Windows.

  • The source package for Linux can be downloaded here: for Linux
  • The source package for Linux can be downloaded here: for MAC OS X
  • The binary of RAREMETALWORKER to run on Linux can also be downloaded here: RAREMETALWORKER BINARY for Linux
  • If you prefer to start from the source files, you can start from decompress the package using the following command:
 tar xvf Raremetalworker.0.4.9.tgz
  • For UM CSG cluster users, no installation is needed. It is available at /net/fantasia/home/sfengsph/code/Rare-Metal/RareMetalWorker/bin/raremetalworker

How to Compile

  • Go to /RareMetalWorker_0.4.9/RareMetalWorker/src and use the following command:
  • If you prefer to use the binary file downloaded above, then no compiling is needed, but it is not guaranteed to work due to system and library requirements.

For compiling questions, please go to FAQ for more information.

How to Execute

  • To execute the program, go to /RareMetalWorker_0.4.8/RareMetalWorker/bin, issue ./raremetalworker.
  • For example command lines, please refer to RAREMETALWORKER EXAMPLES.


Method description and key formulae can be found in RAREMETALWORKER METHOD.

Software Specifications


RAREMETALWORKER is a command line tool. Once you execute, you will see a full list of options printed on the screen.

For detailed description of command options, please go to command reference.

RAREMETALWORKER 0.4.9 -- A Forerunner of RareMetal
          (c) 2012-2014 Shuang Feng, Dajiang Liu, Goncalo Abecasis

 Please go to "http://genome.sph.umich.edu/wiki/RAREMETALWORKER#Where_to_Download" for the latest version.

      Input Files : --ped [], --dat [], --vcf [], --dosage, --noeof
     Output Files : --prefix [], --LDwindow [1000000], --zip, --thin,
       VC Options : --vcX, --separateX
    Trait Options : --makeResiduals, --inverseNormal, --traitName []
    Model Options : --recessive, --dominant
   Kinship Source : --kinPedigree, --kinGeno, --kinFile [], --kinxFile [],
  Kinship Options : --kinMaf [0.05], --kinMiss [0.05]
     Chromosome X : --xLabel [X], --xStart [2699520], --xEnd [154931044],
                    --maleLabel [1], --femaleLabel [2]
           others : --cpu [1]
        PhoneHome : --noPhoneHome, --phoneHomeThinning [100]


RMW needs the following files as input: PED and DAT file in Merlin format, AND/OR a VCF file. When genotypes are stored in PED and DAT file, the VCF file is not needed. However, even if genotypes are saved in a VCF file, PED and DAT files are still needed for carrying covariate and trait information.

PED and DAT Files

  • When PED file has genotypes saved, there is no need for a VCF file as input.
  • RMW takes PED/DAT file in Merlin format. Please refer to PED/DAT format description for details.
  • An example PED file is in the following:
    1 1 0 0 1 1.5 1 23 A A A A A A A A A A
    2 1 0 0 1 1.0 1 34 A C A C A C A C A C
    3 1 0 0 2 0.4 1 43 A A A A A A A A A A
    4 1 0 0 2 0.9 1 13 A C A C A C A C A C
  • The matching DAT file is in the following:
 T YourTraitName
 M 1:123456
 M 1:234567
 M 2:111111
 M 2:222222
 M X:12345
  • DAT file must have variant names in the following format "M chr:pos".
  • Orders of labels in DAT file have to match the order of fields in PED file.
  • Markers in PED and DAT file must be sorted by chromosome and position.
  • Covariate and trait values are saved in PED file. Covariate and trait descriptions are saved in DAT file.

VCF File

  • Another option is to use VCF as input. Please refer to the following link for VCF file specification: 1000 genome wiki VCF specs
  • VCF file should be compressed by bgzip and indexed by tabix, using the following command:
 bgzip input.vcf     ## this command will produce input.vcf.gz
 tabix -p vcf -f input.vcf.gz  ## this command will produce input.vcf.gz.tbi
  • Even with the presence of VCF file, PED/DAT files are still needed for covariates and phenotypes.
  • Are you using PLINK file formats? Converting to VCF is easy. Use WDIST (very similar to PLINK) to make the conversion. Visit this page | WDIST to find documentation and downloads for WDIST.

  • When genotypes are saved in a VCF file, PED and DAT files are used for specifying pedigree structure, covariate and trait information. An example command line might look like this:
 --ped input.ped --dat input.dat --vcf input.vcf.gz
  • When genotypes are saved in the PED file, the VCF file is not needed. An example command line might look like this:
 --ped input.ped --dat input.dat
  • If you want to analyze dosage data from VCF file, the following option has to be specified: --dosage. A key word "DS" in FORMAT field in VCF file has to included accordingly. An example is in the following:
 22	16050408	37239779	T	C	.	PASS	AC=2;AN=496	GT:DS:GP	./.:.:0,0,0	./.:.:0,0,0	./.:.:0,0,0	
 22	16050933	37239784	G	A	.	PASS	AC=141;AN=904	GT:DS:GP	0/0:0.0:1,0,0	0/0:0.0:1,0,0	0/0:0.0:1,0,0
  • --noeof allows using VCF file without BGZF EOF markers. This is a very rare option to use. If your run is terminated with error message: "", then you might want to check out this option.



  • Three files are generated automatically by default:
 prefix.traitName.singlevar.score.txt (single variant summary statistics and QC statistics)
 prefix.traitName.singlevar.cov.txt (covariance matrices of single variant score statistics)
 prefix.singlevar.log (log file)
  • If --zip option is used, then the following will be generated automatically:
  • If --recessive and/or --dominant options are used, then the following files are also generated in addition to the above files
  • If --kinGeno --kinSave is used, then the genomic relationship matrix is stored in
  • If --vcX option is used, then the genomic relationship matrix from chromosome X is stored in


Summary Statistics
  • In the file with summary statistics named prefix.traitName.singlevar.score.txt contains summary statistics that are needed by Rare-Metal. An example is shown in below:
 LDL mean= -0.00, variance=  1.00, heritability= 34.30
 10   45410002        G       A       6103    0.0341589       0.0341589       410     0.165893        126.205 0.309798        4.03074e-10
 19   45412079        G       A       6103    0.0368124       0.0368124       434     0.714645        -265.84 -0.587356       7.87851e-36
 19   45414451        G       A       6103    0.444989        0.444989        5312    0.0759271       -26.1212        -0.00837122     0.640058
  • pvalues from the above output are from the family-based single variant score test.
LD Matrices
  • prefix.traitName.singlevar.cov.txt contains the LD matrix among a variant and the adjacent markers within a prefixed-sized window. The default window size is 1MB. It has the following format:
 CHR    POS        VAR_POS_IN_WINDOW                             LD_MATRIX
 1   762320     762320,865628,865665,878744,879381,1560000    0.0359084,-0.000242112,-0.00125797,-0.000993422,-0.000344509,-0.00017077,
 1   865628     865628,865665,878744,879381,1560000,1864659   0.419804,-0.0103663,-0.00635265,0.0594056,0.0534505,-0.00462183,
 1   878744     878744,879381,1560000,1864659,1877659         0.000404537,-0.000235215,-1.4455e-05,-8.69137e-06,-3.1027e-05,
Genomic Relationship Matrix (GRM)
  • Once --kinGeno --kinSave --prefix options are requested, you would expect to see a GRM generated (compressed by gzip) with name yourprefix.Empirical.Kinship.gz. If --prefix option is not used, then the file name is Empirical.Kinship.gz.
  • If --vcX --kinGeno --kinSave --prefix options are requested, besides the autosomal GRM, you would also expect to see a separate GRM for chromosome X saved (compressed by gzip also) under the name yourprefix.Empirical.KinshipX.gz.
  • The GRMs are generated based on all genotyped individuals included in the PED file; samples with missing phenotype or missing covariates are not excluded from GRMs. This feature makes GRMs reusable if you have multiple traits to analyze in separate runs. You can simple use --kinFile option (--kinxFile option if you have X chromosome GRM together with --vcX option issued) to reuse the pre-saved GRMs.
  • The format for both autosomal and chromosome X GRMs are the same. The first row has all sample IDs (sample size=N) listed. The rest of the file is a symmetric matrix with dimention NxN, and element ij of this matrix represents the kinship between the and the sample whose ID can be found from the first row.
  • For details about GRM calculation, please refer to method.
Log File
  • RMW automatically generates a log file named "yourprefix.singlevar.log".
  • The first part of the log file has options used for your analysis saved.
The following parameters are in effect:

Input Files:
--ped [pheno.ped]
--dat [pheno.dat] 
--vcf [allvars.vcf.gz]
--dosage [false]
--noeof [false]

Output Files:
--prefix [rmw.test]
--LDwindow [1000000]
--zip [false]
--thin [false]
--labelHits [false]

VC Options:
--vcX [true]
--separateX [true]

Trait Options:
--makeResiduals [false]
--inverseNormal [false]
--traitName []

Model Options:
--recessive [false]
--dominant [false]

Kinship Source:
--kinPedigree [true]
--kinGeno [false]
--kinFile []
--kinxFile []
--kinSave [false]

Kinship Options:
--kinMaf [0.05]
--kinMiss [0.05]

Chromosome X:
xLabel [X]
xStart [2699520]
xEnd [154931044]
maleLabel [1]
femaleLabel [2]
  • The second part of the log file has all warnings and running messages saved.
  • RAREMETALWORKER generates QQ plot and Manhattan plots automatically, unless there are only trivial number of variants analyzed.
  • RAREMETALWORKER stores plots of each trait in separate files named yourprefix.traitname.plots.pdf.
  • RAREMETALWORKER stores plots for recessive and dominant results separated with files named yourprefix.traitname.recessive.plots.pdf and yourprefix.traitname.dominant.plots.pdf.
  • RAREMETALWORKER automatically generates three stratified QQ plots, one with all variants, one with variants of maf<0.05, and one with variants of maf<0.01.
  • Genomic controls are automatically calculated and labeled in QQ plots.
  • By using --labelHits option, users can choose to label the hits.
  • Here is an example QQ plot and manhattan plot generated by RAREMETALWORKER.
Single var manhattan.png


  • For special topics such as how RAREMEALWORKER handles missing data, unrelated individuals, markers on chromosomeX, please go to SPECIAL TOPICS.

Example Command Lines

The following list a few popular combinations of options used for analyses. For an itemized description of options, please go to COMMAND REFERENCE.

General Usage

  • If your PED file has many traits but you only want one of them to be analyzed, then the following command does the trick:
 prompt> $PATH/bin/raremetalworker --ped yourInput.ped --dat yourInput.dat --vcf yourInput.vcf.gz --traitName BMI --prefix yourFavoritePrefix
  • If you want to inverse normalize (quantile normalize) your trait before doing associations, this can be done by adding --inverseNormal to your command line:
 prompt> $PATH/bin/raremetalworker --ped yourInput.ped --dat yourInput.dat --vcf yourInput.vcf.gz --traitName BMI --prefix yourFavoritePrefix --inverseNormal
  • The following command will adjust covariates first and then use residuals to proceed association:
 prompt> $PATH/bin/raremetalworker --ped yourInput.ped --dat yourInput.dat --vcf yourInput.vcf.gz --traitName BMI --prefix yourFavoritePrefix --makeResiduals
  • The following command will adjust covariates first and then use the inverse normalized residuals to proceed association:
 prompt> $PATH/bin/raremetalworker --ped yourInput.ped --dat yourInput.dat --vcf yourInput.vcf.gz --traitName BMI --prefix yourFavoritePrefix --makeResiduals --inverseNormal

Related individuals

  • When pedigree is known and you want to use it to count for relatedness then the following command can be used:
 prompt> $PATH/bin/raremetalworker --ped yourInput.ped --dat yourInput.dat --vcf yourInput.vcf.gz --kinPedigree --prefix yourFavoritePrefix
  • When you want to an estimated genomic relationship matrix to count for relatedness then the following command can be used:
 prompt> $PATH/bin/raremetalworker --ped yourInput.ped --dat yourInput.dat --vcf yourInput.vcf.gz --prefix yourFavoritePrefix --kinGeno --kinSave (this will save the genomic relationship matrix for future use)
  • If the genomic relationship matrix has been saved previously, and you want to use it to count for relatedness then the following command can be used:
 prompt> $PATH/bin/raremetalworker --ped yourInput.ped --dat yourInput.dat --vcf yourInput.vcf.gz --prefix yourFavoritePrefix --kinFile yourPreviouslySavedKinship

Unrelated individuals

  • To analyze individuals as unrelated, even if pedigree is known, you just have to use the following command:
 prompt> $PATH/bin/raremetalworker --ped yourInput.ped --dat yourInput.dat --vcf yourInput.vcf.gz --prefix yourFavoritePrefix

Analyzing Chromosome X

  • To analyze markers on chromosome X, if relatedness is not considered, then no special options needs to be issued.
  • When relatedness is modeled using linear mixed model, and pedigree is known, then the following command fits use both autosomal kinship and chromosomeX kinship to fit a variance component model:
 prompt> $PATH/bin/raremetalworker --ped yourInput.ped --dat yourInput.dat --kinPedigree --vcX --vcf yourInput.vcf.gz --prefix yourFavoritePrefix 
  • Adding --separateX to the above command line will only use chromosome X kinship to fit the variance component model:
 prompt> $PATH/bin/raremetalworker --ped yourInput.ped --dat yourInput.dat --kinPedigree --vcX --separateX --vcf yourInput.vcf.gz --prefix yourFavoritePrefix


  • When genotypes are stored in MERLIN format PED/DAT files, command should be the same to do the above analysis, except --vcf option should be excluded.
  • Please refer to PED/DAT format for format requirements.


  • For a comprehensive tutorial of RMW and RAREMETAL using example data sets, please go to the following:

Change Log

  • Version 0.0.1 was released on 11/13/2012.
  • Modified Rare-Metal-Worker to let it output LD matrix by a sliding window. (11/14/2012)
  • Uploaded to public wiki. (11/16/2012)
  • Enabled writing log file by defalut. (11/18/2012)
  • Forced sample IDs to be matched when reading in kinship from a file. Perform a sanity check before reading in kinship file. If a sample of interest is not included in kinship file, then fatal error will occur. (11/19/2012)
  • Added HWE pvalue and call rate in summary statistics output. (11/27/2012)
  • Bugs fixed to solve compiling errors on some machines (Thank you Mary Kate!). Version 0.0.2 released. (11/30/2012)
  • Updated output format. Version 0.0.3 released. (12/3/2012)
  • More messages coded into log file. (12/4/2012)
  • Version 0.0.4 released. (12/5/2012)
  • Bug fixed for empirical kinship calculation when genotypes are read from VCF file. Version 0.0.5 released. (12/6/2012)
  • Version 0.0.6 released. (12/6/2012)
  • Updated output format for monomorphic sites. (12/7/2012)
  • Changed executable name into bin/raremetalworker. Version 0.0.7 released. (12/10/2012)
  • Fixed a bug when reading vcf file with ref or alt allele is missing. (2/5/2013)
  • Fixed a bug when there is missing genotype from VCF file. (2/2013)
  • Fixed a bug when handling chromosome X. Added sex labels option. (3/2/2013)
  • Optimized code to speed up the process of calculating empirical kinship. (3/3/2013)
  • Updated code to report allele frequencies calculated only from selected samples. (3/3/2013)
  • Fixed bug in handling chromosome X. Added sanity checking steps before analysis. Added graphic support by generating QQ and manhattan plots automatically. Upgraded tool to version 2.8. (till 8/12/2013)
  • Added support for analyzing dosages from VCF in version 2.9. (8/27/2013)
  • Fixed the bug which causes crash when writing PDF when all variants are monomorphic. (10/6/2013)
  • Fixed a few bugs handling chromosome X. Generated warning messages when male genotypes are coded wrong in VCF file. (11/25/2013)
  • Released version 0.3.6 and fixed a minor bug that caused by code upgrades from version 0.3.5. (12/4/2013)
  • Released version 0.3.7. Added dominant and recessive models as options. The default model is additive. (1/7/2014)
  • Released version 0.4.0. Added phone home function. Saved Recessive and dominant results in separate files.
  • Released version 0.4.1. Fixed a bug handling variants in nonPAR region on chromosome X when all samples are male.
  • Released version 0.4.2. Fixed a bug that could possibly cause compiling error in some Linux system. Also, in this version, male heterozygous genotypes on chromosome X are considered missing. (3/10/2014)
  • Released version 0.4.3. Fixed a few typo in messages. Added --noeof option for VCF files that does not have a BGZF EOF marker.
  • Released version 0.4.4. Fixed hwe=0.0 issue for monomorphic sites. (3/17/2014)
  • Released version 0.4.5. Fixed a bug when generating plots for recessive results. (3/18/2014)
  • Released version 0.4.6 binary. Fixed a bug in recessive and dominant results. (4/1/2014)
  • Released version 0.4.7 binary and source code. Optimized code to increase analysis efficiency and reduce memory use. Added --separateX option to provide another choice of analyzing chromosome X. Added --kinxFile option to allow kinship X matrix file with different prefix from the autosomal kinship.(4/2/2014)