I was running the cell proportion comparison analysis in my scRNAseq dataset, and when I plotted the boxplot, I could see a few points quite high or low compared to the others. But these are different cases, and I did not find any significant difference across groups. I was suggested by my PI to export the proportion data of each cell type from the individual sample, and remove outliers by the criteria of mean +/- 2SD within the group (i.e. any data point < mean-2SD and > mean+2SD will be removed from that group), then run a regular ANOVA analysis (so that we can find significant difference). I do not think this is a regular way of dealing with RNAseq data, although this may be normally used in other data analysis. And this way in each analysis the outliers will be different. Has anyone done this type of outlier removal before? And what would you suggest if you see this type of data distribution, should I just report no significant difference?
Any suggestions would be helpful, thank you!
