This page contains information about the program called fastNGSadmix which is a very fast tool for finding admixture proportions from NGS data of a single individual and a method for doing PCA of NGS data, using the estimated admixture proporitions. It is based on genotype likelihoods. The admixture estimation part is written in C++ and the PCA part is written in R.
The program + more can be downloaded from:
So far it has only been tested on linux systems.
Input files are genotype likelihoods in the genotype likelihood beagle input file format . Or called genotypes in the binary plink files (*.bed) format  We recommend [ANGSD] for easy transformation of Next-generation sequencing data to beagle format and plink2 for handling plink files.
The example below show how to make a beagle file of genotype likelihood using ANGSD.
HOME$ ./angsd -i example/smallNA12874.mapped.ILLUMINA.bwa.CEU.low_coverage.20130415.bam -GL 2 -sites data/humanOrigins_7worldPops.sites -doGlf 2 -doMajorMinor 3 -minMapQ 30 -minQ 20 -doDepth 1 -doCounts 1 -out example/smallNA12874.mapped.ILLUMINA.bwa.CEU.low_coverage.20130415.bam
Example of a beagle genotype likelihood input file for 1 individual.
marker allele1 allele2 Ind0 Ind0 Ind0 1_14000023 1 0 0.941 0.058 0.000 1_14000072 2 3 0.709 0.177 0.112 1_14000113 0 2 0.855 0.106 0.037 1_14000202 2 0 0.835 0.104 0.060 ...
Where the marker should be chr_pos, instead of rs ID.
A provided SNP.sites file has been included.
The program needs a reference panel of population specific frequencies, populations for which admixture proportions should be estimated, for instance from 1000 G or HGDP, or another custom made reference panel. The program also needs a file telling the size of each reference panel population. There is an R script plinkToRef.R, that can convert a plink file to a reference panel and size of reference panel populations. Example:
Rscript R/plinkToRef.R data/humanOrigins_7worldPops
This generates 3 files, a reference panel named refPanel_plinkFile.txt, a number of individuals file called nInd_plinkFile.txt and a plinkFile.sites file with chr pos and minor major alleles. The .fam file must have the group or population specified as FID, meaning first column and then the individual ID as the second column:
French F1 0 0 1 1 French F2 0 0 1 1 Yoruba Y1 0 0 2 1 Yoruba Y2 0 0 1 1 Yoruba Y3 0 0 2 1 ...
A second argument of MAF threshold can be supplied, meaning all sites where the MAF is below this in any population is removed.
Rscript R/plinkToRef.R data/humanOrigins_7worldPops 0.05
The snpStats R package, is needed for running the plinkToRef.R script.
fastNGSadmix already comes with a premade reference panel, made from Lazaridis et al. (2014) where the curated dataset was selected.
I lifted the dataset hg19 using the program liftOver, I then translated snpNames to rs names, using 1000G data, generating a unique name for each site via "chr-pos-A1-A2" (where A1 and A2 are alphabetically sorted).
Furthermore I removed sites with more than 5 % missing and a MAF below 5 %, and kept only autosomal sites. I selected 7 populations French, Han, Chukchi, Karitiana, Papuan, Sindhi and Yoruba to have representation for most of the world. Furthermore I made sure that I only used unadmixed individuals within each population.
An example of a command running fastNGSadmix with a beagle file and a chosen K of 3:
./fastNGSadmix -likes NA12763.mapped.ILLUMINA.bwa.CEU.low_coverage.20130502.bam.beagle.gz -fname data/refPanel_humanOrigins_7worldPops.txt -Nname data/nInd_humanOrigins_7worldPops.txt -outfiles NA12763_CEU -K 3
If the reference panel has more than K populations, the program will just use the K first populations. You can pick which populations should be analyzed via the "-whichPops" option, where you write the names of the population comma seperated, K has to be same number as selected populations. It should also be noted that the program quits if there are duplicate sites (based on chr_pos ID) in the input or the reference panel.
It then produces two files NA12763_CEU.qopt with the admixture proportions and NA12763_CEU.log.
Or with a plink file (no provided example plink file):
./fastNGSadmix -plink plinkFile -fname data/refPanel_humanOrigins_7worldPops.txt -Nname data/nInd_humanOrigins_7worldPops.txt -outfiles plinkFile -K 3
A whole list of options can be explored by running fastNGSadmix without any input:
The option "-conv" specifies the number of convergence runs, with a new random starting point for each run. This is useful to test for convergence. The program will execute a maximum of 10 times. Another option is "-boot" which specifies the number of bootstrap runs, where random sites are sampled for each run. This is useful for generating a confidence interval of your estimate. The program will execute a maximum of 100 times. For faster inference of admixture proportions the "-Qconv" option can be set to 1, this bases the converge criteria on change in the admixture proportions values. The threshold of this can be set with "-Qtol". By default the "-Qtol" threshold is 0.0000001, it should not be put lower than this and generally this is only for when wanting a fast overview of the data, as it is less precise than the likelihood based convergence. By default the method adjusting the frequencies is used, to use the unadjusted approach set "-doAdjust 0". The accelerated EM algorithm is used by default, to use regular EM algorithm set "-method 0".
After having run the admixture estimation, a PCA analysis can be run using the estimated admixture proportions to account for population structure in the PCA. The PCA method is implemented in R and is run using the script fastNGSAdmixPCA.R.
The method needs the estimated admixture proportions, the analyzed beagle or plink file as well as the genotypes of the reference panel used (as binary plink files). (refPanel is the binary plink files, refPanel.bim, refPanel.fam and refPanel.bed)
Rscript R/fastNGSAdmixPCA.R likes=NA12763.mapped.ILLUMINA.bwa.CEU.low_coverage.20130502.bam.beagle.gz qopt=NA12763.mapped.ILLUMINA.bwa.CEU.low_coverage.20130502.bam.beagle.gz.qopt out=NA12763_CEU geno=data/humanOrigins_7worldPops
Or using analyzed plink files.
Rscript R/fastNGSAdmixPCA.R plinkFile=plinkFile qopt=plinkFile.qopt out=plinkFile geno=data/humanOrigins_7worldPops
By default the populations of the analyzed *.qopt file are used. It can be specified which PCs should be plotted (default PC1 and PC2), by for example giving PCs=2,3 as an argument (if PC2 and PC3). The script also generates barplots of the admixture proportions supplied with confidence intervals.
The snpStats R package, is needed for running this script.