This page is outdated, please also see the man page.

How to run pca?

The QTLtools mode pca allows performing a Principal Component Analysis (PCA) either on molecular phenotype quantifications or genotype data. It is typically used (i) to detect outliers in the data, (ii) to detect stratification in the data or (iii) to build a covariate matrix before QTL mapping.

Phenotype data

In this case, it requires just a single file as input:

A BED file containing the quantification data

To illustrate how this works, we provide the following example file:

A BED file of gene-level RPKM values for 358 samples on chr22: BED / index

Then, to perform PCA on this file, use:


		QTLtools pca --bed genes.50percent.chr22.bed.gz --scale --center --out genes.50percent.chr22

This is going to produce two files:

genes.50percent.chr22.pca that contains the individual coordinates on the Principal Components (PCs)
genes.50percent.chr22.pca_stats that contains the percentages of the variance explained by each PC

If you open genes.50percent.chr22.pca, you'll get something that looks like this:


		SampleID	HG00096	HG00097	HG00099	HG00100

		genes.50percent.chr22_1_1_svd_PC1	9.4607	3.61907	-7.27784	-4.74021

		genes.50percent.chr22_1_1_svd_PC2	-2.89575	-6.1823	-0.488397	-3.1485

		genes.50percent.chr22_1_1_svd_PC3	-3.24429	-4.06885	-6.76097	-0.231029

		genes.50percent.chr22_1_1_svd_PC4	-0.00959694	2.1984	1.58854	-3.01095

		genes.50percent.chr22_1_1_svd_PC5	1.50254	-1.01627	2.46998	4.49168

		genes.50percent.chr22_1_1_svd_PC6	-1.06957	-0.953119	3.60855	2.53457

		genes.50percent.chr22_1_1_svd_PC7	-6.21865	2.10234	-5.26164	3.38665

		genes.50percent.chr22_1_1_svd_PC8	2.62994	-0.516754	0.234968	1.15582

		genes.50percent.chr22_1_1_svd_PC9	-3.62158	0.246679	-8.15425	-1.29294
	
		...

In this file, the header line gives the sample IDs, the first column the ID of the PCs, starting from the first one and each successive line the coordinates of the samples on the PCs. The second output file genes.50percent.chr22.pca_stats looks like this:


		sd	7.81917	7.40651	6.66507	5.40993	4.96369	4.56251	4.41553	3.62532	3.34306

		prop_var	0.100544	0.0902117	0.0730544	0.0481304	0.0405178	0.0342329	0.0320629	0.0216137	0.0183791

		cumm_prop	0.100544	0.190756	0.26381	0.311941	0.352459	0.386692	0.418755	0.440368	0.458747

The 3 first lines give you for each successive PC:

1. The standard deviation
2. The variance explained
3. The cumulative variance explained

The options --center and --scale can be used to enforce centering and scaling of the phenotype values prior to the PCA.

Genotype data

To illustrate how this works, we provide the following example file:

A VCF file of genotypes for 358 samples on chr22: VCF / index

Then, to perform PCA on this file, use:


		QTLtools pca --vcf genotypes.chr22.vcf.gz --scale --center --maf 0.05 --distance 50000 --out genotypes.chr22

The output files are the same than for phenotype data. The only difference here is that we added the two following options to ensure that we have variant sites in linkage equilibrium:

--maf 0.05 to only consider variant sites with a Minor Allele Frequency (MAF) above 5%
--distance 50000 to only consider variant sites separated by at least 50kb

Monday 11th July 2016