Outliers from RNA_seq data

Question

[rajesh]

Hi everyone, I have downloaded the RNA_seq data for PAAD cancer type using TCGA assembler. TCGA assembler also generates a boxplot image, showing outliers. PAAD boxplot for RNA seq data

So my question is this- 1. Is, I have to remove these Outliers sample from my study, as I have to do Differential gene expression analysis. Thanks in advance.

[Kevin Blighe]

Which outlier samples? - the data distributions across this large number of samples look quite similar. A box-and-whisker plot is only part of the story, of course. You should additionally look at a PCA bi-plot of PC1 versus PC2.

[Ram]

Please see How to add images to a Biostars post to add your images properly. You need to use the add image button and the direct link to the image, not the hyperlink button and link to the webpage that has the image embedded (which is what you have used here)

[mikhael.manurung]

Alternatively, if you are using limma or edgeR you can use their robust setting when fitting the linear models so that outlier samples will have less influence on your results.

Answer 1

[GouthamAtla]

Outliers are best assessed from PCA, not from boxplots.

[rajesh]

Thanks.. so instead of looking at Boxplot, I have to do PCA analysis of my data, to remove an outlier.

[Kevin Blighe]

Keep in mind that there is no standard way to statistically define an outlier. If you want, come back here to post your PCA bi-pĺot that you obtain.

[chris86]

Yeah do PCA on all the expressed genes and then on the most variable ones with a few thresholds using co efficient of variation or something similar.