I've been given a large RNAseq dataset on a biological process which I know little about. There is a ton of associated sample data, and I've built a ton of different linear models that include variables that correlate with my process of interest in either the literature or some basic, exploratory analyses.
I'd like to figure out which of these models is worth playing around with while I learn more about the new field. My prior RNAseq analysis experience was all in really simplistic experimental systems, so I'm wondering how to appropriately apply tools like Limma's selectModel AIC/BIC criteria.
Specifically, what is pref? How's it calculated? How should I interpret it?
The two folks that I've spoken to have opposing opinions: One thinks that pref is a broad information criterion score and that I should pick whatever model has the lowest value in this table. The other has told me that the values indicate the number of genes fit well by each model and that I should select the model with the highest value in this table.
As per the manual:
IC
matrix of information criterion scores, rows for probes and columns for models
pref
factor indicating the model with best (lowest) information criterion score
I'd like to figure out which of these models is worth playing around with while I learn more about the new field. My prior RNAseq analysis experience was all in really simplistic experimental systems, so I'm wondering how to appropriately apply tools like Limma's selectModel AIC/BIC criteria.
Specifically, what is pref? How's it calculated? How should I interpret it?
The two folks that I've spoken to have opposing opinions: One thinks that pref is a broad information criterion score and that I should pick whatever model has the lowest value in this table. The other has told me that the values indicate the number of genes fit well by each model and that I should select the model with the highest value in this table.
As per the manual:
IC
matrix of information criterion scores, rows for probes and columns for models
pref
factor indicating the model with best (lowest) information criterion score
Comment