Time Course Analysis With Expression Data

Question

[Floris Brenk]

Hi all,

I recently obtained an expression data set of about 20,000 transcripts of about 80 samples. These samples are from 80 different time points, from 2 weeks to lets say 2 years and the distribution of time points varies a bit. So I know it is possible just to check by simple correlation if the expression of a transcripts increase (or decreases) by time in R (with just cor function). But this function does not allow to take into account variables like RIN value and gender, so does anyone know a more sophisticated method to do this? Also I would like to be able to identify non linear lines for example when a transcript is expressed not 2 weeks till half a year and after 1 year but only expressed from half a year and 1 year old.

Thanks in advance for any help!

[Pavel Senin]

this sounds interesting, I am not sure how your data looks like, but if the series can be labeled (by RIN value, or age and so on) I could offer a technique that I have recently developed targeting time-series characteristic patterns discovery and ranking. It is capable to find class-characteristic subsequences. I guess the best way is to send me an e-mail.

Answer 1

[Charles Warden]

In general, I think the timecourse package in Bioconductor provides the best set of options for analysis of time course data:

http://www.bioconductor.org/packages/2.12/bioc/html/timecourse.html

If I understand the design correctly, I also think linear regression would be suitable (where you explain variation in expression with a model fit using time + RIN + gender).

[Floris Brenk]

He thanks for your reply. I quickly checked the package but did not saw somewhere a function that spits out a list of p-values or that somewhere where you can take into account covariates.

[Charles Warden]

Ok - if the package isn't suitable, then you can just use the standard lm() function for linear regression. This is also probably what I would do, but I know that package can be useful in some circumstances where a simple test (regression, ANOVA, etc.) isn't suitable.