3
$\begingroup$

How can I interpret the PCA results statistically for biological data?

I have used FactoMineR and factoextra libraries for PCA

Scripts used:

library(FactoMineR)

res.PCA = PCA(df, scale.unit=TRUE, ncp=4, graph=F )
par(mfrow=c(1,2))
plot.PCA(res.PCA, axes=c(1, 2), choix="ind")
plot.PCA(res.PCA, axes=c(1, 2), choix="var")
dimdesc(res.PCA, axes=c(1,2))

library("factoextra")
fviz_pca_var(res.PCA, arrowsize = 1, labelsize = 3, repel = TRUE, col.var = "contrib", gradient.cols = c("#00AFBB", "#E7B800", "#FC4E07"))

Here is my PCA output of differnt KO genotypes); how can I explain/interpret the PCA output statistically and if possible biologically: esp, how this colors /contrib explain

enter image description here

$\endgroup$
  • $\begingroup$ Oncidium, did you check it with the wikipedia page on pca or with the help page of the function? Also what kind of data have you used as input (this would help with the biological intepretation)? $\endgroup$ – llrs Jul 22 '18 at 15:56
  • $\begingroup$ @Llopis, This PCA is from targeted proteomics data, sorry I have tagges rna-seq, it should be proteomics. $\endgroup$ – Oncidium Jul 23 '18 at 10:25
4
$\begingroup$

All that plots like this are telling you is that there are some genes that contribute more to the variability seen between your various samples than others. In an ideal world these genes will also be differentially expressed between your groups, but since we don't live in an ideal world that might not be the case. In general, the longer the line and the closer it is to the X axis the more it contributes to the overall variation (since it's contributing more the PC1, which is 53.5% of the variation in your data).

Since you've asked in the comments for a brief "results and description" section style summary of such a PCA I'll include something brief below. Note that one should not over-interpret PCA plots. The axes are (mathematically) orthogonal and don't necessarily correspond to anything biologically relevant.

We performed a PCA on the variance-stabilized counts to check for batch effects and overall clustering of the data. As can be seen, the "3T" and "5T" groups cluster together along the first principal component, while the "0T" and "1T" samples cluster on the opposite side. This suggests that these groups are more similar to each other, as further evidenced by ...

Assuming you had used the vst() function in DESeq2 (I note you've tagged your post with RNAseq, so perhaps you're using DESeq2) then something like that above would be a reasonable description.

$\endgroup$
  • $\begingroup$ Thanks Dr. Ryan, May I ask you if I can have a example of clear description for PCA results. I mean how can I put this kind of results in to words when writing an R&D. I have never seen clear interpretation of PCA results in papers I have gone through. $\endgroup$ – Oncidium Jul 23 '18 at 5:39
  • $\begingroup$ This ends up being rather specific to the topic. In general PCAs are just used for QC to see if there's something odd going on (e.g., an outlier sample or things clustering oddly). I'll update my answer with an description that vaguely fits your image, but be aware that it's best not to over-interpret PCA plots. $\endgroup$ – Devon Ryan Jul 23 '18 at 6:44
  • $\begingroup$ Thanks heaps Dr. Ryan, This helps. I appreciate your kind help. No above PCA is from targeted proteomics data of the sample. $\endgroup$ – Oncidium Jul 23 '18 at 10:11

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.