Differential expression from RNA-seq: variation between replicates

[beans]

Hello,

I’m trying to figure out a method for analysing some RNA-seq data (and I’m very new to this so simple wording please!). I have 2 groups, each with 3 replicates. I want to:

1. assess the variation in expression between 3 replicates, then
2. assess the differential expression between the 2 groups (pooled data from the 3 replicates), taking the into account the variation between the replicates to minimise false positives.

Regarding the first point, I was planning to do combinations of pairwise comparisons to generate DE within the 3 reps (using edgeR or DESeq) but I'm unsure how to then use this to refine the pooled data DE – any ideas or any references where this has been done already? (Or (a) am I totally on the wrong track, (b) is it blindingly simple…?)

Any advice would be greatly appreciated, thanks in advance!

[husamia]

I may have some advice. Can you clarify what you want to compare as "differential expression"?
If you measure expression as read count per mir as I understand to be the case but correct me if I am wrong then you would simply use normalized value per experiment (one sample) then do some type of simple test between the replicates. I did similar thing without replicates. This is one simple easy way that I used but there are many other ways. I was just trying to see if I can find any differences.

[kwatts59]

Use the cufflinks and cuffdiff software to determine differential expression.
Here is the link

http://cufflinks.cbcb.umd.edu/

[chadn737]

Both edgeR and DESeq are made to work with replicates, so I don't see why you need to do a pairwise comparison of each replicate before looking for differential expression between your two conditions. Just specify which samples are replicates when you enter the data into edgeR and DESeq and you should be fine.

[beans]

Hi All,

Thanks very much for your comments - I was hoping it would be fairly simply and that those packages would take replicates into account. It's difficult to anticipate these things first time round while you're waiting fo the data to come in!

Husamia, in response to your question, I have 3 replicate samples from each of 2 species and I want to assess DE between the individuals of each species to get an idea of what kind of variation occurs within a sample of the population Then I want to assess DE between the 2 species overall to find out which genes are most markedly different, but using the between-replicate variation to minimise false positives. Hope that makes sense. I'm new to the packages that calculate DE and from what I read it seems they calculate DE from pairwise comparisons, hence my thoughts that that was necessary.

Cheers again,

[husamia]

Thanks for explaining I like to clarify to myself. Is the reason for the replicates to account for the bias created by the enrichment and other experiment bias? or they are different replicates? I want to clarify this point. I have concern in my data that there may be false positive DE caused by amplification bias. In other words, I am concerned that I may be detecting preferential PCR amplification instead of real expression of certain amplicons by the nature of the enzymatic approach and depending on sequence of amplicons. Just curious thought.

[beans]

Hi,

Sorry for the delayed response,I didn't realise there had been another reply to this. The replicates are simply different individuals that we wanted to sequence in order to get an idea of the level of intra-specific variation in gene expression. Unfortunately, this hasn't been possible - the quality of the RNA was such that individuals had to be pooled Maybe next time! I hope your analyses have gone well

Cheers,
Beans