I have a PCA plot from DESeq2's plotPCA(vsd, intgroup=c("conditions")) function. I have 9 samples distributed in to 3 groups of 3 biological replicates each. My reason for trying to visualise all on the same PCA plot is to see that samples from the same group (condition) cluster together and to see a separation between the groups.

The PCA plot I have is does not appear as I expect:

1-Can I just remove a sample which I deem is the outlier and perform a reclustering of the rest on the PCA?

2-What parameter can I use to declare that a point is an outlier rather than an in-group Biological variation?

3-I also wish to find out if it is allowed to plot more than two groups of sample at a time on a PCA plot.

Thank you in advance for your kind hints.

PCA plot

  • 1
    $\begingroup$ Rather than doing any automated analysis, I like to do some exploratory analysis in these cases. Plot the counts of some individual genes from which you know that they should change and some that should not change. Also see if normalization seems appropriate on the global scale, e.g. by producing pairwise MA-plots between all the samples and see if the majority of data points is well-centered at y ~ 0 and if plots with this outliers look odd. Also try to increase the number of genes for PCA. Try to get an idea if this sample is abnormous. $\endgroup$ – ATpoint Mar 20 at 10:22

Since you have so few points here, it is hard to tell which are outliers by just looking at the PCA plot. I suggest you use correlation heatmap for your gene expression and check whether there are obvious outliers.

As for using more than one groups when plotting PCA, I think you can just try passing a vector with many groups to the argument intgroup.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.