For an experimental design that the control and treatment samples are not paired and where n= 4-5, I wonder what is the best way to measure the variability. My second question is how actually p-values are calculated to show whether fold change is significantly up or down, comparing a fold change value of 1? I am using error propagation to measure the variability, but is there any other way that is well accepted? By the fold change i mean the concentration level of treatment group divided over the concentration level of control group, for a given metabolite.
if you want to calculate and modeling experimental designed data, i suggest you to use from MINITAB16 software . because its very simple user inference and very powerful to create and calculate designs,
now how to calculate p-value manually; p-value is probability of accept hypotheses test if test is true. so its variable face and formula target by target..
Hi Ebru,
I'm not sure how things are done in the metabolomics literature, so this might be a little non-canonical. My answer will contain some R code, so I hope you're at least somewhat familiar.
Suppose we have two populations x and y:
n
If the control and treatment samples are both normally distributed then u can use t test or ANOVA to calculate p value .
I would search for some literature where metabolomics fold changes are reported. I think it might be usual to report the log2 fold change, rather than the absolute fold change. When the variance between replicates is much higher for the higher measurements than the lower ones, then the log transformation will stabilise this variance.
Hi Ebru,
I do exactly similar experiment and use volcano plots with p-values calculated using Welsh t-test(as your data do not follow normal distribution) on y-axis, and fold change of x -axis(Using non-parametric test is also an option, the results wont get much better but number of statistically significant differences will get lesser).You will need at least 5 biological replicates to increase the power of the test. If you take 3 samples, you only have 2 degrees of freedom and this hinders. I think having 5-6 replicates is a must, having even more is a plus.
If you use R regularly, than its wont take you long. If not then you can use this online webtool.
metaboanalyst.ca
You may try Progenesis from Waters/Nonlinear:
http://www.nonlinear.com/progenesis/qi-for-proteomics/
Fold change calculation is one of its nice features.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis