Question: Differences among phenotypes with different subgroups
gravatar for bwassie
15 months ago by
bwassie10 wrote:

Hi all,

I am working on an ATAC seq data set for which I have two phenotype groups (control and disease). Within each of these groups there are different cell types specific to the groups as shown in the experimental design:


Phenotype cell type
1 Control control1
2 Control control1
3 Control control2
4 Control control2
5 Control control2
6 Control control2
7 Control control3
8 Control control3
9 disease disease1
10 disease disease1
11 disease disease1
12 disease disease1
13 disease disease2
14 disease disease2
15 disease disease3
16 disease disease3

I am looking for a way to run differential binding analysis that will model the different cell types separately (instead of modeling all controls together it will build separate models for control1, control2...etc) and find differential sites between control and disease. This is because there is a lot of variation across the different cell types for each phenotype (eg. there is high variation between control1 and control2 cell types).

I have tried using DiffBind using the cell types as blocking factors but that gave me an error (DBA_FACTOR=Phenotype and DBA_TREATMENT=cell type):

dba.contrast(dba_count, categories=DBA_FACTOR, block=DBA_TISSUE, minMembers=2)

Warning messages:

1: Blocking factor invalid for all contrasts: 2: No blocking values are present in both groups

I think this may be due to the cell types being specific to only control and disease phenotypes. Is there a way to redefine the groups so as to take advantage of blocking for cell type without losing biological relevance?


I have also looked at edgeR as it seems to have an example in the user guide (section 3.5) that seems to be similar to what I am doing (hence why I tagged it here) but they re-number the different patients in that example and I'm confused as to why they did that.

If you can help me with any part of these questions I would appreciate it!




ADD COMMENTlink modified 15 months ago by Rory Stark2.5k • written 15 months ago by bwassie10
gravatar for Rory Stark
15 months ago by
Rory Stark2.5k
CRUK, Cambridge, UK
Rory Stark2.5k wrote:

DiffBind only explicitly supports a "blocked" design, where specific samples in one group share a factor value with samples in the other group. Do the control1 and disease1 cells share something in common (as well as control2/disease2 and control3/disease3)? If so you can use the group (1, 2, or 3) as the value and you will get the effect you desire. If control1 and disease1 samples don't share anything in common compared to the other cell types,, you would have to model that manually in edgeR or DESeq2.



ADD COMMENTlink written 15 months ago by Rory Stark2.5k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.2.0
Traffic: 214 users visited in the last hour