[BioC] Methods for time course gene expression analysis in an observational cohort

[Oliver Hofmann]

Dear all,


posting this on behalf of a colleague who is looking for help with a 
time series analysis:

"I've read a number of articles and searched the list's archive, but 
cannot figure out what is the best method to analyze time course data 
where I need to adjust for confounders.

First - my experiment: Longitudinal cohort study in 80 children (60 
asthmatics and 20 non-asthmatics), three time points (one prior to 
exposure, two after exposure), one exposure (endotoxin) and several 
measured confounders (other allergens, asthma vs. no asthma, atopy vs. 
no atopy).

The research question is:

1) Are there differentially expressed genes in response to endotoxin at 
the two time points after exposure (early and late) - note issue of 
confounding exposures

2) Do differentially expressed genes differ between asthmatics and 
non-asthmatics at the two time points in response to endotoxin exposure

Most time course studies are laboratory studies where by experimental 
design there are no (known) confounders.  Limma or a linear mixed 
effect model seem to handle time course with covariates.

Can anybody give me advice on:

1. What is the best method (limma vs. linear mixed effect model vs. 
other) for this research question and design, and _why_ one would be 
preferable over the other?

2. What is the best way to decide which covariates to include (prior 
biological knowledge vs. algorithm)

3. How does one typically handle missing data in this kind of time 
course microarray studies?"

Any feedback would be appreciated!

Best, Oliver


-- 
Research Scientist    Harvard School of Public Health
Associate Director    Bioinformatics Core
Skype: ohofmann       Phone: +1 (617) 365 0984
http://compbio.sph.harvard.edu/chb/