Is plotMDS (from limma package) a PCA (Principal Component Analysis) or a PCoA (Principal Coordinate Analysis)?

Question

[harelarik]

A. Is the plotMDS function (from the limma package) based on PCA (Principal Component Analysis) or a PCoA (Principal Coordinate Analysis)?

B. What the values on the plot represent? Is it root-mean-square average or root-mean-square deviation (see below)?

• In the limma paper published in Nucleic Acids Res at 2015 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4402510/) it is stated that: "Distances between samples on the plot represent ‘leading fold change’, which is defined as the root-mean-square average* of the log-fold-changes for the genes best distinguishing each pair of samples. "

• In the R manual it is stated that: "The distance between each pair of samples (columns) is the root-mean-square deviation (Euclidean distance) for the top ‘top’ genes. Distances on the plot can be interpreted as _leading log2-fold-change_, meaning the typical (root-mean-square) log2-fold-change between the samples for the genes that distinguish those samples."

Thank you,

Arik

[prasundutta87]

This video may be helpful for a basic understanding on this topic.

[harelarik]

Sorry but the link does not work.

[ATpoint]

Should work now.

Answer 1

[Gordon Smyth]

A. The plotMDS documentation help(plotMDS) explains in detail that's doing multidimensional scaling (MDS) and PCoA is just a synonym for MDS. PCA is a special case of PCoA so in some cases plotMDS is also doing PCA. If the axis labels says "Principal component" then it is PCA (and PCoA). If the axis label says "leading logFC" then it's just PCoA.

B. Both expressions mean the same thing, it's the root-mean-square of the largest logFCs: sqrt(mean(logFC^2)).

[harelarik]

Thank you very much! A. How can I make the function generate PCA? So the axis label will show "Principal component".

Arik

[Gordon Smyth]

gene.selection="common"

[harelarik]

Thank you very much!

As common option will "select the same genes for all comparisons" when would you recommend to use it?

[Gordon Smyth]

I recommend you use the plotMDS default settings, unless your dataset is very large (hundreds or thousands of samples) in which case gene.selection="common" is much faster.

[harelarik]

Dear Dr. Smyth,

Thank you very much!

Arik