Weird MAplot with DEGs lying along the diagonals

Question

[FedeXandeR]

Dear BioStarrers, I'm currently reprocessing some (not so) old microarray data originally acquired using the following platform: Agilent-026652 Whole Human Genome Microarray 4x44K v2 comparing a tumor sample and the matched "healthy" region nearby, in the same biopsy.

I'm interested in differential gene expression, as usual, but I get really weird MAplots (and quite compressed BoxPlots) in the preprocessing step. Here is an example:

https://ibb.co/LN1rptj

https://ibb.co/n336Kv6

https://ibb.co/DK52MFT

https://ibb.co/3R0PzxD

https://ibb.co/mb3KZxK

https://ibb.co/271T7xw

It seems like the typical bulk of the highly-expressed-unchanged genes is completely missing, while all the DEGs have been "pushed" towards the two diagonals of the plot... Never seen something like this before... Do you think this is biologically reliable? Maybe some problem with hybridization or scanning?

Any help/suggestion/comment from anyone more experienced than me is welcome!

Thank you so much!

I think the problem is in the biology or in the wet-lab step, and not in the coding. However, here is the (very essential and almost standard) code I used:

library(limma)
targets = readTargets("Targets.txt", row.names = "SampleNumber")
raw = read.maimages(targets, source = 'agilent.median', green.only = TRUE)
   # Here I generate some QC plots of Raw data
raw_BGcorrected = backgroundCorrect(raw, method = "normexp", offset = 50)
   # Here I generate some QC plots of BG-subtracted data
raw_BGandNormalized = normalizeBetweenArrays(raw_BGcorrected, method = "quantile")
   # Here I generate some QC plots of Normalized data

[Kevin Blighe]

Cross-posted: https://support.bioconductor.org/p/9135484/

[FedeXandeR]

Sorry, I post so rarely that I didn't know cross-posting was an issue... do you think I should remove one of the two?

[jared.andrews07]

It's fine, we just want folks to be able to find the answer if it is answered elsewhere. If you get an answer there before here, please add the answer here.

[FedeXandeR]

Oh, ok, got it! I'll do it. Thank you!

Answer 1

[seidel]

Your unnormalized expression plots have 50% of your data at only 64 counts. That seems really low. If you examine histograms of the background and foreground signal in each channel, do they look different? Do you have a data set based on a successful hybridization that you can use for reference, for what distributions of expected values might look like? Also, Agilent has various hybridization controls that you can look up - have you checked out the values of those?

[FedeXandeR]

Thank you for your accurate advice. Actually I suspect a general failure of chip hybridization, as also suggested by Gordon Smyth on Bioconductor Forum where I cross-posted the same issue: Bioconductor cross-post. By the way, the analysis of Agilent control probes seems to confirm that. Thank you again bye