RIP-Seq data analysis -- unsure which #'s to compare!

[sjeschonek]

Hello SEQAnswers -- I'm hoping you can provide some insight.

We recently performed a RIP-seq experiment in our lab and I'm a bit unsure about how best to analyze the results. We are very new to RNA seq and any advice would be helpful.

We are essentially looking for the overlap of RNAs between two RIPs, minus any background against an IgG control.

For example, each experiment contained:
IP with Antibody-A beads
IP with Antibody-B beads
IP with IgG beads

We are interested in those genes that appear in both A and B IPs but not in the negative IgG control.

We have 3 biological replicates for all IPs.

I used the Trinity-based program, Agalma (http://www.ncbi.nlm.nih.gov/pubmed/24252138) to assemble my reads.

I currently have some quantitative data, including expression counts and FPKM.

I'm having trouble in determining what is background and what is significant. I had originally planned to compare FPKM across samples. For instance, if the Gene X had an FPKM score of 50 in the IgG, I would only consider Gene X a significant hit if it had an FPKM score >50 in both Antibody-A and Antibody-B IPs.

What is concerning me is that one of the control genes which we KNOW should be highly enriched by the A/B IPs shows a much higher FPKM expression in the negative IgG control than it does in the A/B IPs (ie: 170 vs 80). These FPKM values are consistent across all replicates.

This makes me think that perhaps comparing FPKM across samples is not the best method and may not be valid. Should I be using a different means of comparison?

Does anyone have experience or insight to identifying background in RIP-Seq experiments? Should I be using different values or a different calculation?

Thank you VERY much for any help. It's greatly appreciated.

[sjeschonek]

I notice this thread has a good number of views but no responses. Is there a better forum/subforum to post this question on? Thanks!

[dpryan]

While I've done a few RIP-seq experiments, I've not done one with your design before, so what follows will just be a first thought.

One simple, but sub-optimal, approach would be to use any of the standard tools (DESeq2, edgeR, etc.) to test for differential expression between the Antibody-A or Antibody-B vs. IgG samples and then take the intersect of the results. This is less than ideal since you'll miss genes that are on the margin of significance in one or both samples. However, it'd be a good starting point and would give you somewhat conservative results for the validation step.

I can think of some more involved methods that would (at least partly) be more powerful, but they'd be a bit more involved and possibly not worth the effort (give what I mentioned above a try, if you don't get much of anything then post back here).