I am currently using the MetaboAnalyst 6.0 website and have a few questions about the statistical options for the t-test and volcano plot.
For both the t-test and volcano plot, if I select the Limma option, is Limma using the data file after it has already been normalized, transformed, and scaled in MetaboAnalyst? For example, I am using normalization by sum, log10 transformation, and auto-scaling.
I also wanted to clarify what is meant by “small sample size” in the context of the Limma function. In general, what sample size range would be considered small for Limma analysis?
For the t-test option, MetaboAnalyst states that Limma “works on log2-transformed normalized data for better performance on small sample sizes.” What does this mean, especially if my data are already log10-transformed in the preprocessing step?
For the volcano plot using moderated Limma, I wanted to confirm how the axes are calculated. Does the log2FC from Limma on the x-axis represent the estimated log2 fold change calculated from the Limma model? Also, does the y-axis represent −log10 of the moderated p-value from Limma? Does the CSV file reflect this calculation? I noticed that the volcano plot for classical mode also uses log2FC and -log10(p-value), and the description says that it uses classical FC (raw data) for the x-axis and Student’s t-test p-values for the y-axis. What is the major difference between these two (classical vs moderated limma)?
Lastly, when presenting the t-test and volcano plot results, is it best to use the same statistical method for both? For example, if I use moderated Limma for the volcano plot, should I also use the Limma t-test option so that the p-values and significant metabolites are consistent across both analyses?
Any clarification would be greatly appreciated. Thank you!
Small sample size
There is no hard threshold. My experience is that over 12 samples (concentration table), classical t-tests and limma give very similar results. Less than 12 samples, limma gives more stable results
Normalization for limma
Limma works on log-transformed data. We perform log2 after sample normalization (i.e. after adjusting tissue weight, sample volume, batch, dilution factor, etc). Your log10 normalization or scaling does not affecting the results
Limma and volcano plot
If you choose limma, both p-values and logFC are from limma
Using t-test and limma-based volcano plot results
Totally up to you.
The view and result files should be consistent. However, some documentation may not up to date
You can compare values in the graph (mouse over) and check the values in the downloaded file;
From the R history, you can locate the R functions in the MetaboAnalystR. If you are not familiar with R, ask AI to explain it.
Hello Xia.Lab, thank you for the clarification! Sorry, I have a few more questions.
For the Limma:
When the log2 is done after the sample normalization (in my case, the normalization by sum), and that is done for both the t-test and the FC?
For the volcano plot, is the FC, log2(FC), p-adjusted, and -log10(p) used from the data that has undergone sample normalization + log2 (limma)? Is the log10(p) and log2(FC) applied after the log2 transformation and sample normalization?
If my log10 normalization and auto-scaling don’t affect the results, and I want to use limma for both the t-test and volcano plot, then when do the log10 data transformation and auto-scaling (data scaling) come in play?
Even if I do not choose the log2 data transformation in the data normalization page in MetaboAnalyst 6.0, does selecting the Limma option for the t-test and volcano plot automatically perform the log2 after the sample normalization (normalization by sum)?
Thank you!
If my log10 normalization and auto-scaling don’t affect the results, and I want to use limma for both the t-test and volcano plot, then when do the log10 data transformation and auto-scaling (data scaling) come in play?
No. limma has its own normalization (log2)
Even if I do not choose the log2 data transformation in the data normalization page in MetaboAnalyst 6.0, does selecting the Limma option for the t-test and volcano plot automatically perform the log2 after the sample normalization (normalization by sum)?
Thank you for the clarification. I had another question, which is for the PCA plot. Would the PCA plot be utilizing the sample normalization (normalization by sum), log10 data transformation, and auto-scaling (data scaling)?