I performed a microarray study of 60 samples. 20 samples for each of Group A and B, C. I used the "samr" package to look for differentially expressed genes between subgroups and calculate the power.
On the comparison of Group A vs. C, SAM identified 1123 genes differentially expressed at FDR = 0.01 (1000 permutations). Given the definition of FDR, the number of truly differentially expressed genes could be anticipated as 99% of the total number of genes identified. Consequently, assuming 1112 genes truly differentially expressed, the power of tests reach 0.94 and the type I error equal 0.01.
On the comparison of Group B vs. C, SAM identified 1094 genes differentially expressed at FDR = 0.01 (1000 permutations). Similarly, assuming 1083 genes truly differentially expressed, the power of tests reach 0.92 and the type I error equal 0.01.
Is it correct to interpret the results in this way?
On the comparison of Group A vs. B, no differentially expressed gene was identified by SAM at FDR = 0.01 (1000 permutations). In this case, how to interpret the outcome of samr.assess.samplesize (attached figure)? Furthermore, how can i use the observation of no differentially expressed gene to demonstrate that Group A and B are more similar in a statistically significant way?
Many thanks for your help!
Best regards,
Qinghua XU
You can use Principal Component Analysis with your group A, B, C using BABELOMICS (http://babelomics.bioinfo.cipf.es/) in utilities, to see the similarity of your groups.
When you are doing a t-test comparing differential expressed genes with a lot of genes, you have to calculated the adjusted p-value because you are improving the error in each gene comparison. I think, your graphic is showing this...the power of FDR is decreasing when you are comparing more genes.
In Babelomics you have to used the t-test with a Benjamini and Hochberg (BH) or FDR that is considering this error.
Hello
I also performed microarray in past. We used the software SAM and it was easy to use. You can perform ANOVA and other statistical analysis with much ease when compared to SAMr or other software. If you have some know how about excel it is much easier to perform t-test etc
you're statements are a bit too definitive. you made an effort to control the false discovery rate at 1% and observed the results. does that mean 99% of those genes are truly differentially expressed? no. you're working with real data (and real noisy data), where normality was violated (often) and independence is truly not the case.
as Patricia mentioned clustering should demonstrate the similarity between groups A and B. variance is another option, but a venn diagram should suffice.
Do NOT trust the statistics in Excel.
Why don't you simply compare the genes that are significant for group A vs. C and B vs. C? If there is an overlap this inidcates a similarity. You can do this simply using Venn diagramms (http://bioinfogp.cnb.csic.es/tools/venny/index.html).
As to Patricia's comment, SAM is a bootstrapping approach to multiple testing correction, you do not need further post-tests.
Thanks, Ulrich. Indeed, the genes that significant for group A vs C and B vs C are 80% overlapped.
Hi Patricia, while working with SAM and FDR criteria, the multiple testing issue has been taken into account.
Correlate W-Curve sequences with microarray expression maps of Mab interation.
Remember Causality does not come from one correlation
Day 9 : CommanderHotdougCork: and the last of the Planet Lost Redog1 on Mars
- Badges
- Science topic
- Similar topics
- Bioinformatics
- Bioinformatics and Computational Biology