RIP-seq : normalize Input with Immunoprecipitated raw counts or not

Question

[am835821]

EDIT: OP deleted most of the content of their post just retaining the two sentences below, but I've restored the original content from Google Cache.

- Ram

---

Hi everyone, Thanks for your time and your help !

---

ORIGINAL CONTENT:

---

0 6 months ago am835821 • 0

Hi everyone,

I am analyzing a RIP-seq experiment made of 12 RNA libraries as follows :

6 "control" libraries : 3 input (total RNA) and their corresponding immunoprecipitated RNAs (IP) and 6 "affected" libraries : 3 input (total RNA) and their corresponding immunoprecipitated RNAs (IP)

• Input control 1 - Input control 2 - Input control 3
• IP control 1 - IP control 2 - IP control 3
• Input affected 1 - Input affected 2 - Input affected 3
• IP affected 1 - IP affected 2 - IP affected 3

I would like to analyze on one hand the 3 Input control vs the 3 Input affected and on the other hand the 3 control IP vs the 3 affected IP.

I am starting with a single raw count table of the 12 libraries. My question is : Should I split the table in half at the very beginning, an Input count table and IP count table and then perform all the normalization and DE analysis steps in parallel ? Or should I keep the 12 libraries in the same count table and perform normalization on the whole ?

I tried both and outputs are different unless I'm mistaken. I can not figure out which is the relevant choice. In my opinion, it is not suitable to compare the 6 Input between them from a count table normalized with the whole 12 libraries.

Thanks for your time and your help !

[ATpoint]

I have no experience with RIP-seq but usually in something like ChIP-seq one only uses inputs for peak calling as they are globally different from the actual IPs so proper normalization is close to impossible for a DE analysis. Is this the case in RIP-seq?

[am835821]

Thanks for you reply, I have no experience in ChIP-seq and I am just starting the analysis workflow so I don't really know if RIP-seq Input are handled the same way as it is in ChIP-seq but maybe I can precise what we would like to get if it helps :

• We expect from the inputs libraries to identify wehter they are deregulated genes between control and affected individuals (mutated for a specific protein binding RNAs).
• And we expect from the IP libraries to show if the RNAs bound to this specific protein (mutated or not) correlate somehow with the deregulated genes identified in the Input ones.

I assume that the Input and IP libraries should be normalized for raw counts separately but if I do it that way, none of the adjusted P values are significant whereas when I normalize on the whole table with both Input and IP libraries, I have like 50 significant deregulated genes.