Then, we combine this data (as shown below) with the metadata to use different aesthetics and colors on the plot.

From the above object, to get the scatter plot for the samples, we need to look into vst_pca$x. Now, let us look into building a plot out of these components. So, looks like the first two components explain almost 85% of the data. So, one should look into the structure of the PCA object and import it into ggplot accordingly! Note There are quite a few functions in R from different packages that can run PCA.

rotation: the matrix of variable loadings (columns are eigenvectors).

x: the coordinates of the samples (observations) on the principal components.

sdev: the standard deviations of the principal components.

gc_vst <- read.table('data/counts_vst.txt', header = T, row.names = 1, sep = ' ')Īfter you computer the PCA, if you type the object vst_pca$ and press TAB, you will notice that this R object has multiple vecors and ames within it. If we do not transpose, then PCA is run on the genes rather than the samples.

Therefore, we transpose our count matrix using the function t(). It takes in a matrix where samples are rows and variables are columns. To run PCA, we use the R function prcomp(). For this, we will use the VST data, because it makes sense to use the normalized data for building the PCA.

Pca column 64bit 4shared