Time Course Rna-Seq Analysis

Question

[kanwarjag]

I have a data with three replicates for 4 time points and underlying aim is to find signature genes for a particular phenotype which will occur after 3rd time point and goes down at 4th time point. I was wondering how can I analyze this, Should DEseq package (count based ) method be of some use I have not seen any time course study with RNA-seq using Deseq. Any other suggestion or pointers will be helpful.

Thanks

[Dave Bridges]

are you testing the hypothesis that there is a non-zero change in expression (slope) or that the slopes differ over time between two groups?

[kanwarjag]

My aim is to find out the genes which are specifically associated with particular phenotype over time. Yes in other word genes which are nonzero over time course

Answer 1

[Dave Bridges]

You should be able to use fitNbinomGLM and nbinomGLMTest from the DESeq package:

library(DESeq)
cds <- makeExampleCountDataSet()[ 1:100, ]
cds <- estimateSizeFactors( cds )
cds <- estimateDispersions( cds, method="pooled" ) #need to use either pooled or blind
time.course <- c(0,2,5,15,24)
fit1 <- fitNbinomGLMs( cds, count ~ time.course ) #fits to the formula that the time.course is a covariate
fit0 <- fitNbinomGLMs( cds, count ~ 1 ) #the null hypothesis
nbinomGLMTest( fit1, fit0 ) #chi squared test for each gene

I am not 100% sure, but i think you will also need to adjust the results of nbinomGLMTest for multiple observations.

If you need a time course plus a condition then just modify the formula to add that variable

Answer 2

[Charles Warden]

If you use log2(RPKM + rounding factor), you can use the timecourse package:

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

This would generally be my preferred strategy (with rounding factor = 0.1). However, it looks like you only care about changes between the 3rd and 4th timepoint. If this is the case, you only need to define two groups (pre and most) and any differential expression package could probably be used. If the sensitivity is too high (because the "pre" group is so large), you could use a combination of strategies: 1) compare 4rd versus 3rd timepoint (using DESeq, for example) and 2) filter for genes not differentially expressed between 1st to 3rd timepoints (using timecourse)

Answer 3

[ashutoshdhingra]

I am in a similar situtation like you. For differentiation expression during time course control vs mutats. Can you tell me which Bioconductor package and script did you use for DE? I have neural differentiation time line of control vs mutants (2 replicates for 5 time points).

Your reply will be highly appreciared.

Best regards, Ashutosh