I performed Pearson correlation coefficient calculation (PCC) on the gene expression data and then on the principle component space (considering all components). The PCC looks different in both the cases. Why?
Hello Jayesh,
This is because the PCA performs an orthogonal transformation of your Y original variables (that may have some degree of correlation) in Z non-correlated variables. The Z linear uncorrelated variables are your Principal Components. Those differences that you found are caused by a property of the PCA calculation (I would expect absence of correlation between PCs. You probably found values very very close to zero, for example: 1e-25, which is ~0)
However, you still may use the results of PCA (transformed data) to interpret the results found in Pearson's correlation (original data). For example: there is a possibility to find that high correlated variables in your original data are responsible for an interesting pattern of variation in your samples, and you may see this pattern of variation using the PCA.
It is worth to mention that the text above is a general statement about how PCA works in a set of hypothetical variables. I am not sure how those analysis would behave in a gene expression experiment.
I hope this answer may help you. Let us know if you have any questions about it.
Best wishes,
João Paulo
Since PCA extracts linearly independent components from the data, correlation coefficients calculated on Z variables are free from collinearity.
Hi João Paulo,
I am not comparing results from PCC analysis and things I found in PCs.
I am trying to find co-expressed gene pairs for which I am calculating Pearson correlation coefficient (PCC). For calculating PCC, I tried using log2 Fold change values and also co-ordinates from the PCA space after PCA transformation.
PCC values calculated using PCA space coordinates were really higher than the PCC values calculated using Log2 fold change values. Why?
Jayesh
- Badges
- Science topic
- Similar topics
- Biological Science
- Genetics
- Gene Expression