minimac is a low memory, computationally efficient implementation of the MaCH algorithm for genotype imputation. It is designed to work on phased genotypes and can handle very large reference panels with hundreds or thousands of haplotypes. The name has two parts. The first, "mini", refers to the modest amount of computational resources it requires. The second, "mac", is short hand for MaCH, our widely used algorithm for genotype imputation.
A full release of
minimac is expected here soon. In the meantime, a binary only Linux version of minimac is available from here for those who are willing to test pre-release software. The current version should be stamped 2011.08.18 - if your version shows a different version number or date stamp when it runs, it is not current.
If you use this beta version, please be sure to stop by the MaCH download page and fill out the registration form, so that we can let you know when an official release is available and keep you updated with respect to any bug fixes.
The current version of minimac comes in two flavours,
minimac-omp. The latter version uses the OpenMP protocol to support multi-threading, resulting in faster throughput.
Questions and Comments
For the pre-phasing step, cost increases quadratically with the number of states and linearly with the number of rounds. The following table provides a simple example.
|States||Cost per round|
|100 states||3 min|
|200 states||12 min = (3 min * 22)|
|400 states||48 min = (3 min * 42)|
|500 states||75 min = (3 min * 52)|
So, in this case running haplotyping with 500 states and 10 rounds would require 75 min * 10 = 750 min.
Typically, haplotype quality improves rapidly with the number of states but only slowly with the number of rounds. We recommend running ~20 rounds of the MaCH haplotyper and selecting a number of states as high as your patience will allow (but ideally greater than 200).
A good rule of thumb is that minimac should take about 1 hour to impute 1,000,000 markers in 1,000 individuals using a reference panel with 100 haplotypes. Performance should scale linearly with respect to all these factors. So, your approximate computing time in hours should be about:
These statistics refer to a single core in a modern Intel CPU core and, although your mileage will vary, most modern CPUs should be no more than a few times faster (or slower) than that.
If you are estimating model parameters at the same time as imputing missing genotypes, you can account for the time needed for parameter estimation with the following formula:
In this updated formula, Nrounds represents the number of iterations used for parameter refinement and Nstates represents the maximum number of reference and target haplotypes considered for each update.
Using minimac for genotype imputation involves two steps. First, you will have to estimate haplotypes for your entire sample -- this will be the more computationally demanding step. Once that is done, you will be ready to quickly impute missing genotypes using the reference panel of your choice.
Estimating Haplotypes for Your Sample
For the haplotyping step, we current recommend using MaCH with the --phase command line option. As input MaCH will need Merlin format pedigree and data files. All markers should be ordered according to their physical position and alleles should be labeled on the forward strand.
Preparing Your Data
Within each file, markers should be stored by chromosome position. Alleles should be stored in the forward strand and can be encoded as 'A', 'C', 'G' or 'T' (there is no need to use numeric identifiers for each allele).
We recommend that, if at all possible, you should phase your chromosomes according to NCBI build 37. Future releases of the 1000 Genomes Reference panel and other public sets of reference haplotypes are expected to use this genome build.
If figuring out position and strand for each marker seems like hard work, don't despair. For you, this should be the hardest bit of the entire process! For the computer, the fun is about to start.
NCBI build 36 / NCBI build 37
The 1000G June reference panel is on build 36, future 1000G reference panels will be on build 37. This has also some impact on the pre-phasing: some SNPs will ordered differently in each genome build (we expect order will be more accurate in more recent builds!). Purists will claim that imputation using NCBI build 36 and NCBI build 37 reference panels requires phasing using the exact same reference panel -- others will claim this makes little difference.
A typical MaCH command line to estimate phased haplotypes might look like this:
mach1 -d sample.dat -p sample.ped --rounds 20 --states 200 --phase --interim 5 --sample 5 --compact
This will request that MaCH estimate haplotypes for your sample, using 20 iterations of its Markov sampler and conditioning each update on up to 200 haplotypes. A summary description of these parameters follows (but for a more complete description, you should go to the MaCH website):
||Data file in Merlin format. Markers should be listed according to their order along the chromosome.|
||Pedigree file in Merlin format. Alleles should be labeled on the forward strand.|
||Number of haplotypes to consider during each update. Increasing this value will typically lead to better haplotypes, but can dramatically increase computing time and memory use. A value of 200 - 400 is typical.|
|| Iterations of the Markov sampler to use for haplotyping. Typically, using 20 - 30 rounds should give good results. To obtain better results, it is usually better to increase the |
||Request that intermediate results should be saved to disk periodically. These will facilitate analyses in case a run doesn't complete.|
||Request that random (but plausible) sets of haplotypes for each individual should be drawn every 5 iterations. This parameter is optional, but for some rare variant analyses, these alternative haplotypes can be very useful.|
||Tell MaCH to estimate phased haplotypes for each individual.|
||Reduce memory use at the cost of approximately doubling runtime. This option is recommended for most GWAS scale datasets and computing platforms.|
Imputation into Phased Haplotypes
Imputing genotypes using minimac is an easy and straightforward process: after selecting a set of reference haplotypes, plugging-in the target haplotypes from the previous step and setting the number of rounds to use for the model parameter estimation, imputation should proceed rapidly.
Creating SNP List File
Minimac requires a file listing markers in the haplotype file. This file can be easily generated by extracting the second column from the .dat file. In a standard Unix system, a command like this should do:
cut -f 2 -d " " sample.dat > target.snps
A typical minimac command line might look like this:
minimac --refHaps ref.hap.gz --refSnps ref.snps.gz --haps target.hap.gz --snps target.snps.gz --rounds 5 --states 200 --prefix results
A detailed description of all minimac options is available elsewhere. Here is a brief description of the above parameters:
||List of SNPs in the reference panel|
||Reference haplotypes (e.g. from MaCH download page)|
||SNPs in phased haplotypes. These should largely be a subset of the SNPs in the reference panel.|
||Phased haplotypes where missing genotypes will be imputed.|
||Rounds of optimization for model parameters, which describe population recombination rates and per SNP error rates.|
||Maximum number of reference (or target) haplotypes to be examined during parameter optimization.|
||Optionally, a string that is used to help generate output file names.|
Reference haplotypes generated by the 1000 Genomes project and formatted so that they are ready for analysis are available from the MaCH download page. As of this writing, the most recent set of haplotypes are based on genotype calls were generated in May 2011 and are an interim analysis of Project's Phase I data.
Imputation quality evaluation
To evaluate imputation quality, Minimac hides data for each genotyped SNP in turn and calculates 3 statistics:
- looRSQ - this is the estimated rsq for that SNP (as if SNP weren't typed).
- empR - this is the empirical correlation between true and imputed genotypes for the SNP. If this is negative, the SNP is probably flipped.
- empRSQ - this is the actual R2 value, comparing imputed and true genotypes.
These statistics can be found in the .info file
Additional Sources of Information
If the combination of MaCH and Minimac still runs too slowly for you, and you have access to a multi-processor compute cluster, you can look at ChunkChromosome page to learn how to conveniently split each chromosome into multiple segments that can be analyzed in parallel.
If you are especially interested in 1000 Genomes Imputation, then you should look at the Minimac: 1000 Genomes Imputation Cookbook.
X Chromosome Imputation
minimac supports the imputation of genotypes on the X chromosome (non-pseudo-autosomal part).
- Split the X chromosome pedigree file by sex.
- For females: follow the same protocol as for autosomes (phase with MaCH and impute with minimac).
- For males
- Remove the pseudo-autosomal part (for build hg18: chrX:1-2709520 and chrX:154584238-154913754 ; for build hg19 chrX:60001-2699520 and chrX:154931044-155260560)
- Convert the pedigree file into a MaCH haplotype file (missing genotypes should be encoded as: "0" or "." or "N" ) and impute using minimac as described above.
- <Example of a male only pedigree file>
- FAM1003 ID1234 0 0 M A/A A/A C/C
- FAM1004 ID5678 0 0 M 0/0 C/0 G/G
- <End of pedigree file>
- Note that, consistent with the Merlin convention, hemizygous males are listed as if they were homozygous.
- <Example of the corresponding haplotype file>
- FAM1003->ID1234 HAPLO1 AAC
- FAM1003->ID1234 HAPLO2 AAC
- FAM1004->ID5678 HAPLO1 0CG
- FAM1004->ID5678 HAPLO2 0CG
- <End of the corresponding haplotype file>
Post-imputation Association Analysis
Please use mach2qtl.
Please use mach2dat. Versions 1.0.18 and above accommodate to minimac output.
If you are looking to learn about small computers made by Apple, Inc., you have come to the wrong page. Try looking at http://www.apple.com/macmini/, instead.
If you are looking for a low calorie version of the Big Mac sandwich, you'll be sad to know the Mini Mac has been discontinued. However, you are not the only one who likes the idea of a Mini Mac and you'll probably find some company on the web .