Hello everybody,
I have some gene expression experiments (real time PCR arrays) with many groups with biological replicates, but with a very small sample size (3,6,4,3,3,3; and non paired). I have no trouble calculating fold changes, but since fold change is just a change on magnitude and does not entail statistical significance per se, a statistical test is necessary for most publication purposes.
Due the aforementioned small sample size, it's almost impossible to assume a distribution and hence a parametric test like a paired T-student test does not sound too feasible to me. I've been thinking in something non-parametric, such as a Mann-Whitney Test, but I think that achieving a good enough p-value would be almost impossible. I've also thought about trying to relate confidence intervals with fold-change or something like that.
Any ideas? How have you dealt with this situation?
Thanks in advance.
Regards.
1) non-paremetric tests make no sense particularily for small sample sizes.
2) dCt values are well-known to have an approximate normal distribution, so there is actually no reason to doubt that this assumption makes sense.
For PCR arrays you will have information about several (many) genes, what allows you to get quite good estimates of the variance. You should have a look at "limma" (https://www.ncbi.nlm.nih.gov/pubmed/16646809; https://bioconductor.org/packages/release/bioc/html/limma.html). It focusses on microarray data, but you can use dCt values instead of log(spot intensity) values. I used it for the analysis of qPCR data and it worked very well.
I remmebered that the Support Vector Machine (SVM) or relative improved algorithms are suitable for handling small samples, you can have a try.
Good luck.
Also, there is another algorithm named SMOTEBoost, is resampling and for expanding small groups of sample to a large numbers of sample, hope that can be helpful.
Thank you all your help. I think I'll try Limma, as Jochen kindly suggested.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics