Question: Time-series or pairwise comparison EdgeR - what's more suited?
0
gravatar for Biogeek
2.6 years ago by
Biogeek340
Biogeek340 wrote:

I've been doing a bit of reading regarding my experiment and I'm currently deciding what approach to take - I know that stats should be planned prior to experiments; however, I'm having a brain wobble.

I set out and completed a 2 time point RNA-seq analysis for 2 conditions. Control for each time point and a treatment.

I have two mind-frames:

1./ Two-way comparison - EdgeR classic Given I have two time points - a control for each; apply an EdgeR classic approach and do pairwise comparisons between conditions at individual time points. Why? We know organisms have different life stages and their regulation changes dynamically. The control samples of each time point are independent separate individuals from one another. They are not related. I feel that time-series analysis in this situation may be assuming too much stays constant?

2./ Apply an EdgeR GLM complex model for multi-factorate analysis.

My findings so far. If I conduct the classic approach, my BCV's are lower by doing inidivudal time points, than combining all conditions in a complex GLM. When I compare everything in the GLM, there are not many diff. expressed genes between comparisons on each day (Ctl Vs treat day 1) - (Ctl Vs treat day 6).

What I'm ultimately asking is, what the appropriate approach for this sort of set-up?

Thanks.

ADD COMMENTlink written 2.6 years ago by Biogeek340

The strategy to takes depends completely on the question you want to ask of the data.

ADD REPLYlink written 2.6 years ago by Devon Ryan88k
1

I'm not sure two timepoints make a series.. Nonetheless, I'd probably download some existing data where two time points are available with +/- treatment at each point: and decide how best to analyse your data on this... rather than optimise your analysis on your own data to suit your required output from that data

ADD REPLYlink modified 2.6 years ago • written 2.6 years ago by russhh4.2k

No disagreement on whether two times constitutes a time series :) As far as I can tell the question boils down to, "I have a 2x2 factorial design, should I use that design or do pairwise comparisons?"

ADD REPLYlink written 2.6 years ago by Devon Ryan88k

I'm gonna say the appropriate thing to do is stick to a 2x2? My BCV is sitting at 0.5 :-( Does that mean my data is awful?

ADD REPLYlink written 2.6 years ago by Biogeek340

Well..if you say two different life stages, that doesn't sound so bad. See disucssion section of http://nar.oxfordjournals.org/content/40/10/4288.long

ADD REPLYlink written 2.6 years ago by fanli.gcb660
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1080 users visited in the last hour