merging RUV-corrected datasets in DESEq
1
0
Entering edit mode
Jakub ▴ 30
@jakub-9073
Last seen 8 months ago
United Kingdom

Dear Bioconductor Community,

I have a workflow for multiple plates single-cell RNASeq that works as follows

  1. QC
  2. Use RUVSeq to eliminate library complexity bias from data (this works really well, with pre and post)
  3. Account for Zero Inflation using Zinbwave
  4. Model differential expression using DESeq with model:  ~W_1 + group

I appreciate any comments on this, but this works very well, with improved ERCC estimates (better fit) and increased detection of known differentially expressed genes (two transgenes).  (This is not the question)

I have multiple plates, and each is biologically different from the next (different age, region), however all have condition X and Y. My collaborator wants to merge some analysis. I am not sure that combining these plates is biologically entirely valid, but my collaborator insists that there is biological justification for this. This was obviously not planned, otherwise better batch design would have (possibly) overcome that. (Again not the question)

What I would like to therefore do is combine the plates. This is obviously possible at multiple stages, but I presume the following approach would be best:

  1. QC plates independently
  2. Use RUVSeq to obtain W_1 for each plate
  3. Run zinbwave on each plate
  4. Run a model that accounts for the batch such as: ~ group + age

Now using this approach, I have two or more W_1 terms that I don't know where to put? Obviously there may be alternative approaches that are better.

Many thanks,

Jakub

deseq2 glm zinbwave ruvseq • 381 views
ADD COMMENT
0
Entering edit mode
@mikelove
Last seen 20 hours ago
United States

I've asked about this recently, and I don't think it's so simple how to combine comparisons across biological replicates. I can see many ways to go, but I don't think there is software ready to do this just yet. There are technical and biological differences in comparing A vs A' (whether conditions or cell types) across biological units. I could imagine either assuming a fixed effect across biological units, or a random effect, etc. DESeq2 can only model fixed effects.

Maybe the ZINB-WaVE team will have more experience as far as how to normalize across multiple plates.

ADD COMMENT

Login before adding your answer.

Traffic: 659 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