I downloaded a dataset from GEO (GSE1906) and they have only 1 sample (I mean with out any replicates/triplicates) per condition and their data files contain the measure of the abundance of a transcript (which I assume as gene expression). I would like to know how to calculate the statistics (p-value, FDR) in order to identify differentially expressed genes if we have only 1 reading.
Sorry to disappoint you but one cannot make statistics when n=1. Any attempt to do this would be meaningless and misleading. Analysis of isolated samples may provide some hints and basically allows to generate hypotheses. But to check these hypotheses one imperatively needs cross-evidence or replicates.
From my own experience, I can say that fold changes obtained in isolated RNA-seq studies where n=1 for both compared groups are often misleading. The impact of stochasticity is quite important: expression of some genes may dramatically change as a result of any external factor affecting this particular sample, without you even being aware of it. This source of variability is random and has nothing to do with your experiment. Using replicates alleviates a part of this stochasticity simply because it is not reproducible. And only reproducible things will stay - the ones you are actually looking for!
Hi Alexander,
Thanks for the reply. Yes, i am aware of all the above mentioned problems and hence not making any predictions right now from the data. But one of my colleague told me that we can somehow predict the uncertainty (as he heard somewhere). I will hunt if i can find some.
cheers,
Aravind
For affy arrays with both mis-match and perfect-match probes, you can use raw probe signal (i.e. non-normalized intensity data) to derive statistical significance between single pairs of arrays. See the BioConductor library "SScore" and the companion paper:
"Empirical validation of the S-Score algorithm in the analysis of gene expression data. By Richard E Kennedy, Kellie J Archer1, and Michael F Miles. 2006. BMC Bioinformatics 7:154" (and references therein).
In the situation with no replicates (admittedly undesirable, but sometimes unavoidable as for example, in cases of human tissue samples), I've found this approach to be useful to look at pairwise arrays.
This method will soon be released as a public package. There are ways to use information across samples even if none of them are replicates. Always better to get replicates though!
http://arxiv.org/abs/1301.4144
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Bioinformatics
- Bioinformatic Tools