specific design and model

[farmohh]

Hello, Thank you very much for the great tool! I have eCLIP dataset with two conditions (before and after treatment), 8 IP replicates plus SMI controls in replicates. I would like to use DEWSeq to compare binding profile of a RBP at two different time points. I was wondering if the DEWSeq pipeline can be applied to search for differentially bindind regions between IP samples (not for a one-sided IP vs SMI comparison)? If so, how can I make this comparison while accounting for the negative controls? What would be the design formula and model for this experiment?

Here is the sample info:

Sample ID Condition1 Condition2 1 T0 A IP 2 T0B IP 3 T0C IP 4 T0 D IP 5 T2 A IP 6 T2B IP 7 T2C IP 8 T2 D IP 9 T0 A Input 10 T0B Input 11 T0C Input 12 T0 D Input 13 T2 A Input 14 T2B Input 15 T2C Input 16 T2 D Input

The different condition are T0 or T2 (non-stimulated or stimulated).

So for example, “T0 A IP” is the non stimulated sample A with the IP “T0 A input” is the same sample as before just the input

“T2 A IP” is the stimulated sample A with the IP “T2 A input” is the same sample as before just the input

Thank you in davance! All the best,

[sudeepsahadevan]

Hi @farmohh

Thanks for using DEWSeq. If I understand your question, you would like to test whether there are any differentially bound regions between "T0 IP" samples vs "T2 IP samples", and you are not looking into testing between "T0 A IP" vs "T2 A IP", please clarify this.

If you are looking to test between "T0 IP" samples vs "T2 IP samples", we would recommend not to disregard Input samples totally, but use them as a part of your design. The first strategy would be a 3 step analysis, that is

1. T0 IP vs T0 Input
2. T2 IP vs T2 Input
3. Compare 1 and 2 and find ids of regions that are common, not found in any This would be a simple binary comparison to see if any of T0 enriched regions are present in T2 enriched regions.

The second strategy would be to test the ratio of ratios as it is done here. This should be feasible as DEWSeq uses DESeq2 for statistical testing. We haven't tested it ourselves as we didn't have such a study design before, but we are happy to follow up/help with this.

Hope that helps.

Kind regards, Sudeep.

[farmohh]

Hi Sudeep, Thank you very much for your kind answer, very helpful indeed. yes, I want to see whether there are any differentially bound regions between "T0 IP" samples vs "T2 IP samples" and would love to go with the second strategy (the ratio of ratios).

So my colData looks as such: Sample id assay condition patient 1 IP pre A 2 IP pre B 3 IP pre C 4 IP pre D 5 IP post A 6 IP post B 7 IP post C 8 IP post D 9 SMI pre A 10 SMI pre B 11 SMI pre C 12 SMI pre D 13 SMI post A 14 SMI post B 15 SMI post C 16 SMI post D

my design is therefore: ~ assay + condition + assay:condition As I have paired samples, I was wondering whether the sample should be treated as a random variable and be included in the design matrix as bellow or the first design is sufficient? design = ~ patient + assay + condition + assay:condition

And I really appreciate your willingness to help us. I was wondering if I can share some data with you via email or I shall continue asking my questions here? Thanks in advance.

[sudeepsahadevan]

Hi @farmohh

My initial intuition is that this design: ~ assay + condition + assay:condition will be better as you have at most 2 patients per condition when you use this design design = ~ patient + assay + condition + assay:condition

I would have to stress at this point that we haven't done this before, the data we had so far were from simple case vs control experiments, so you might run into some issues. If you are willing to share data with us, we gladly welcome that. I'm assuming that the data is some your in-house experiments, so if you'd like to share the data, either you can post a sample of it here or use the contact email which is: biohentze[at]embl[dot]de

Kind regards, Sudeep.