HI, everybody! Before sequencing all samples we want to be sure that we can identify several differentially expressed genes between two initial samples: one is like reference (or wild-type) the other is experiment. Also I should metion that reference sample have a much more reads than the experiment sample (30 millions against 5 milions). What statistical test would be appropriate in this situation?
Totally agree with everything Michael said. Best thing to do would be to repeat the experiment with more replicates and at a greater sequencing depth. With modern sequencers you can achieve a reasonable depth in multiplexed reads with just a few lanes and that isn't actually all that expensive these days.
If cost or effort, however, is an issue you can compare the samples by fold change and rank the genes accordingly. Go for the candidates with the strongest change and try to find corroborating evidence in online databases / literature. Then rigorously test the candidates in small-scale experiments (PCR & co) in the lab and be prepared to encounter a lot of false positives (and also to miss true positives, but of course you wouldn't know).
Michael B Black, thank you for your answers! After I got them, I relaxed a little)) Finally, I used Kal's et al. Z-test on proportions that is suggested for situations like mine in CLC Genomic Workbench. I don't know how accurate are results, but I could find upregulation in the expession of genes that should be upregulated in our model. I totally agree with you about replicates, they are coming! And also I have ERCC spike in control kit, I'm going to use it next. Do you use it? If yes, then when do you add it: to total or polyA-RNA?
I would also suggest you not to deal with the statistical significance, because one can not determine whether the expression differences were due to experimental or biological noise, when no replicates are present. In such case, it is probably better to describe the expression differences only in terms of relative measurements, like fold change or FPKM differences.
But in some experimental designs, biological replicates are not possible. By assuming that the mean is a good indicator of expression variance, and only a minority of genes are affected by the condition, we can indeed estimate the statistical significance. Software packages such as DESeq2 (http://www.bioconductor.org/packages/release/bioc/html/DESeq2.html) and NOISeq (http://www.bioconductor.org/packages/release/bioc/html/NOISeq.html) do support the estimation of significance when no biological replicates are present.
- Badges
- Science topic
- Similar topics
- Biological Science
- Microbiology
- Microorganisms
- Bacteria