Batch effect identification using PCA

Question

[mohammedtoufiq91]

Hi,

I have a question about batch effect. I used the qRT-PCR transcript targeted assay with around 300 genes with 160 samples (included 30 samples from batch 1 and 130 samples from batch 2). Batch 1 experiment and Batch 2 experiment was performed on different days. I used normalized log2-scale relative expression value (Delta Ct= Geometric mean of Housekeeping genes - Gene of interest) in R to plot the Biplot PCA by making use of library("ggfortify"), library("FactoMineR"), library("factoextra"). I have attached a screenshot of the plot. B_1 is batch 1 and B_2 is batch 2. Does this show batch effect? If yes, then how could this be handled. What should be the % of PC1 and PC2. Could you please share any useful resources to understand batch effect and batch correction procedure.

Thank you,
Toufiq

Answer 1

[Kevin Blighe]

Hey,

There is no way for any of us to say whether there is a batch effect [that requires correction] or not just by looking at this. I would be interested in determining why those yellow samples at the top group away from the main bunch. Try to determine the origin of those and it should enlighten you.

Evidence against there being a batch effect is illustrated by many blue and yellow samples grouping together.

Kevin

[mohammedtoufiq91]

Thank you for the inputs @Kevin Blighe. I will look the annotations for the yellow samples that grouped at the top.

Regarding this, Evidence against there being a batch effect is illustrated by many blue and yellow samples grouping together, if the samples are grouped that means it has batch effect? or sample should form different clusters based on batches? Thank you again.

[Kevin Blighe]

if there existed a batch effect, then I would expect B_1 and B_2 to cluster away from each other, i.e., non-overlapping.

[mohammedtoufiq91]

Thank you for the comments @Kevin Blighe.