(posted from our now obsolete bitbucket issues)
Would the following dataset be possible for a DIABLO analysis??
The dataset comprises of the following characteristics 1. Two-group analysis: a control vs drug-treated. 2. Each group contains 3 time-point treatment (treatment time is identical, i.e., 0, 1, 12hr) 3. At each time point, 3 omics (RNA, DNA, Protein) measurements were obtained.
There are different types of analyses you can do, assuming you have enough samples per time point and per group. (enough == more than 3!)
Option 1: you are primarily interested in a two group analysis
First: assess the effect of time, and whether the individual variation (unique individuals) is greater than the effect of time. This can be done with a simple PCA on each data set, and compared with a PCA(argument multilevel = sampleID) to see whether you see any differences in the plots. In the former, if you see that individuals clusters irrespective of time, then you have a strong individual variation, which should be removed with the multilevel approach. If you do not see a strong individual variation, carry on.
Then, depending on Step 1, you can transform your data to accommodate for the individual variation by applying the withinVariation() function and extract the ‘within variation’ data (see ?withinVariation example) and input into DIABLO where Y = control vs drug.
Option 2: you are interested in trends across time, as well as control vs drug
Have a look at our latest article in press, (bioRxiv version) and specifically Figure 2 to see what kind of question you can answer. However, you will need to analyse each group separately.