For LMMS modeling for timeOmics, does your values for each gene get converted into the fitted values based on the selected model? For example, I have log10 transformed values for proteomics that are changed after being fit to a linear model with LMMS for some genes, and once doing a z-score of those genes they result in the same z-score values and are linear (due to the low variance of those transformed values). Is that due to the fitted values from a Linear model, spline, etc.? And is it fine to proceed with those type of results? Also for modeling with LMMS, I’m using just the average of 2 replicates for each timepoing (average → log10 → LMMS → Z-score by multiblock.pls)
Hi @bzavala,
does your values for each gene get converted into the fitted values based on the selected model
Yes
See my earlier response regarding log10 transformation, it might not be appropriate
. But first, the LMMS should include as many replicates per time point as possible. n=2 (if I understand correctly) is on the lower / impossible end to be fitted with a LMMS, and I am not sure how the LMMS would be running in yourur case (it should throw an error if you run it on an average, e.g one value per gene per time point across the whole sample set). From the LMMS fitting you should not transform further (ie following the vignette of timeOmics is sufficient).
I hope that helps, I feel I am missing crucial details to answer.
Kim-Anh
For further context, I am trying to integrate two types of transcriptomics and a proteomics dataset. So 3 datasets taken from the same 7 timepoints. The transcriptomics datasets consisted of 2 replicates per timepoint and was then normalized across all replicates. The transcriptomics replicates were then averaged for each timepoint, however the proteomics consisted of only one replicate per timepoint. After filtering for low abundance in every dataset, I next transformed the values to log2 (transcriptomics) and log10(proteomics). The reason for this is that we wanted to reduce the variance that is caused by using the untransformed values and it becomes very useful for selecting relevant genes when doing feature selection. I next used timeOmics with LMMS of every gene on the averages replicates and was successful in doing so (no errors). Is it necessary to run timeomics on the replicates instead of the average? I’ve noticed that some of the examples provided from timeOmics do not specifically state the manipulation of replicates, if they do, could you please direct me to that example (the example I have been using is this timeOmics_HMP_T2D_seasonality/seasonality.Rmd at master · abodein/timeOmics_HMP_T2D_seasonality · GitHub)?