Coverage for duplicates

[billstevens]

Hi guys,

So I'm doing a 100bp PE reads on the human genome with 3 different conditions. I ran one set on one lane, and I just finished looking at the data. The plan originally was that after I looked at the data, if it looked like there were good differences (there are), I would run two sets of samples on one more lane.

However, now that the moment of truth is here, I'm wondering if this is the right move. Should I instead run only one more set of data for one lane, and just use my results in duplicate?

Sorry, I'm a such a newbie, so I'm not sure what stats are pertinent so let me just give what I have. The RNA is very good, and in the set that I have results for, I got 90% read alignment. After I used cuffdiff, cummeRbund gave me 300 significantly differntially expressed genes. So 30 of the 300 had "values" under 1. However, of the 300, about 40 had over a two fold difference, and of these 40, 25 of them had a "value" under 1.

For microarrays, a two fold change is the minimum you can have to call it useful. If that's the rule on sequencing, then I worry that if I split up my lane in essentially half, I'll lose 25 of the 40 genes that were very significantly differentially expressed.

Any thoughts or suggestions?

Thanks so much!

[billstevens]

So I had some more info if anyone was searching and came across this. First off, so the values are FPKM values (I had a deadline yesterday, and I was running around all crazy and didn't actually stop and THINK).

So to address this problem, see how many mappable reads you have. I had approximately 100 M per condition. If I use two sets of samples on lane, I'll end up with 50 M per condition. With an FPKM value of 1, I'd have 50 fragments. I spoke to an Illumina Tech who told me they hear in the 30-50 range as the minimum in which you can still use. Is this what everyone else has heard?

Someone else seemed to wonder this as well:

http://seqanswers.com/forums/showthread.php?t=9776&highlight=minimum+FPKM