Looking for perspective on a odd MA plot / skewed data
0
0
Entering edit mode
22 months ago
jevanveen ▴ 20

Hello Biostars!

I (and collaborators) recently performed a patch-seq experiment... That is, we did patch-clamp recording of neuronal firing, then aspirated the contents of recorded cells, and made RNA seq compatible libraries.

Data look alright - certainly there is a lot of PCR duplication. Probably a combined effect of too many PCR cycles and no UMIs in the kit we used.

To analyze, I did a STAR -> deduplicate -> htseq-count -> DESeq2 pipeline. I think I am getting meaningful results and cells that are expected to be similar cluster together.

But I have an odd MA plot that is giving me some pause:

MA plot

The low expression transcripts are DRAMATICALLY skewed toward one group. I am thinking that this must be an artifact. In my downstream analyses, I have filtered based on a minimum of normalized counts, but I am wondering if anyone has encountered an odd graph like this and how you have dealt with it? Should I filter raw reads before performing DESeq? Thanks very much in advance!

RNA-Seq patch-seq • 1.2k views
ADD COMMENT
1
Entering edit mode

Can you show the unshrunken MA-plot? This is apeglm right? How many replicates is this?

ADD REPLY
0
Entering edit mode

Thanks for responding!

Replicates are 3 in one condition, 4 in the other.

Sorry, do you mean MA-plot with non-shrunken LFC values? I am still running my alignments and DESeq on usegalaxy. As far as I can tell, the LFC values are not shrunken by default.

ADD REPLY
0
Entering edit mode

Yes that is what I meant. This here is most likely a shrunken MA plot. Unshrunken plots have highest logFC at low average expression while shrunken ones tends to shrink low expressed genes towards LFC of 0 (this is from my limited understanding at least the basic idea of shrinkage in this context). I would try DESeq2 in R outside Galaxy to have better control over all parameters. It is computationally not expensive, every laptop can do it. Doing so you can produce the unshrunken MA-plot.

What you can also do is to post this question over at the Bioconductor support forum so that the DESeq2 maintainer (Mike Love) can have a look. Still I would try via R on a local machine first so that you can post exact code.

ADD REPLY
0
Entering edit mode

This is helpful information... I have a decent amount of experience with R, but I did my first seq analyses on galaxy, and so... if it aint broke, don't fix it, right? Well, I guess now it's broke, lol. Will give it a shot. Cheers

ADD REPLY
0
Entering edit mode

You can follow pretty much 100% the example code for DESeq2 from the manual for standard analysis. It is just a few lines of code.

ADD REPLY

Login before adding your answer.

Traffic: 2683 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6