Hello,
I am truly struggling. I have RNA raw counts data that i am analyzing with DEG and GSEA. Came up with a pipeline and everything, but now that I want to clean the data and run my pipeline.. I AM STUCK.
For quality check i use VST (blind) to make PCA, RDA, dendrogram, and Distance heatmap. Also, using Tukey's IQR method. However, nothing is consistent and my RDA Pvalue is all >0.05. I have 6 sample groups (2speciesX3tissues) and all of them have N from 10 to 20 max, mostly unbalanced between control and treatment.
I was told by my supervisors that the data can be "exploratory", but i still want to have clean samples and I am not sure how/if i should remove any samples, especially because the ones i remove end up changing the genes tested. How would you approach this?
0 answers
No answers yet.
Log in to answer this question.
Do you observe similar replicates and sample groups together during your quality check step?
sometimes treatment/control cluster all together and sometimes they are sparse. depends on specieXtissue.
Show us, and even better your supervisors, the PCA etc. If the replicates don't cluster, then you won't get much out of the data. You do have replicates, right ?
i do have around 7-11 biological replicates. My current PCAs are from VST(blind), which is a new thing im trying based on some older posts here. These are 2 examples where my gut tells me to just remove them, especially if they correlate to low reads. However, i don't know if they are outliers per say or extremes? Also, i hate using arbitrary cutoffs, like what even is a low read, i've been using 5M.
There might be some QC issues with your RNA-seq data. Some things to check are (i) the number of mapped reads should be >15 million for human. (ii) The proportion of reads mapped to the transcriptome/genome should be >80%. (iii) The proportion of ribosomal RNA should be less than 10%. This could explain some of the patterns you are seeing in the PCA and give you justification for excluding some low quality samples.
If there are no QC issues, then there could be other sources of variation including batch effects or even differences in the tissue and cell type composition. For example a muscle sample may include different amounts of leukocytes, fibroblasts, and occasionally connective tissue. Generally, these can be identified and corrected for in the downstrream DE analysis.
Lastly, there may not be any differentially expressd genes in your system due to lack of replicates or sequence depth or the lack of biological differences.
In some cases although no individual genes reach statistical significance, pathway tools like GSEA can still give useful information on subtle changes.