How to cluster samples with PCA in R?

[rozitaa]

Hi,

I have RNA-seq data for 16 mouse samples. I would like to cluster cufflinks results of these samples by PCA in R. The data looks like this:

gene_id    gene_short_name    FPKM_101    FPKM_102    FPKM_103    FPKM_104    FPKM_105    FPKM_106    FPKM_107    FPKM_108    FPKM_109    FPKM_110    FPKM_111    FPKM_112    FPKM_113    FPKM_114    FPKM_115    FPKM_116
uc007aeu.1    -    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0
uc007aev.1    -    0    0    0.0095358    0.0095358    0.011704    1.48E-10    2.05E-63    0.0083273    0.014457    0.0068505    0.0053635    0.022235    0.0047757    0.018794    0    0.01661
uc007aew.1    -    0    0    0    0    0    1.2568    0.27389    0    0    0    0    0    0    0    0    0
uc007aex.2    -    0    0    0    0    0    7.1538    0.0096687    0    0    0    0    0    0    0    0.0050925    0
uc007aey.1    -    8.27E-07    0.00049201    0.00043141    0.00043141    0.00079353    0    0.00074324    1.56E-09    3.39E-20    1.16E-61    1.80E-20    1.72E-09    1.56E-96    5.13E-07    5.34E-07    4.78E-07
uc007afb.1    -    0.40549    2.08E-248    1.11E-19    1.11E-19    2.40E-93    0    0.49777    0.10711    1.22E-12    0.014644    6.48E-13    0.11777    0.02695    0.25169    0.08951    0.080144
uc007afc.1    -    1.93E-06    0.38845    0.34061    0.34061    0.31315    0    0    5.01E-09    1.04E-19    0.00046243    5.53E-20    5.51E-09    0.17202    1.20E-06    1.16E-06    1.04E-06 

Commands in R:

data=read.csv('raw_cuff_data.csv', header=TRUE)
data_pca <- prcomp(data[, 3:18]) 

How should I plot them in order to see clustered samples. I know my pca data are in data_pca$x, but how should I cluster them? and plot one point for each sample?

Thanks

[dariober]

Hi,

I have RNA-seq data for 16 mouse samples. I would like to cluster cufflinks results of these samples by PCA in R. The data looks like this:

How should I plot them in order to see clustered samples. I know my pca data are in data_pca$x, but how should I cluster them? and plot one point for each sample?

Thanks

Hi - Try something along these lines...

## Test data:
dat<- matrix(data= rnorm(n= 10000), ncol= 10)
colnames(dat)<- paste('sample_', 1:ncol(dat), sep= '')

## PCA
pcaResult<-prcomp(t(dat))

## Set up plot
plot(pcaResult$x,
    main= 'Principal components of samples',
    xlab= sprintf('PC1 (sd: %s%%)', round(100 * (pcaResult$sdev[1] / sum(pcaResult$sdev)))),
    ylab= sprintf('PC2 (sd: %s%%)', round(100 * (pcaResult$sdev[2] / sum(pcaResult$sdev)))),
    type= 'n'
)

## Plot labels
text(x= pcaResult$x[,1], y= pcaResult$x[,2], labels= rownames(pcaResult$x), cex= 0.5)

Samples similar to each other will group together in the plot but keep in mind that PCA doesn't do any clustering.

Dario

[rozitaa]

Thanks, a lot. I also transposed my data and I got the exact pca plot that I wanted. I just wasn't sure if that's the correct way.

[Jeremy]

There is a nice PCA generated in the DESeq2 vignette that you could look at and apply to your cufflinks data.