Hi, 

I have been trying to learn deseq2 and managed to work on a two factorial design (followed a similar post DESeq2 Model Design and worked through my analysis, thanks to Michael). I'm now into a three factor time-series design and would need some help please in designing my model and describing contrasts. I have been looking into lots of forum questions but still confused, have also read the vignette.   

Here is my design: Group: A and B, Treatment: control and treated, time: 0,1,2,4 - 3 replicates each

group treatment timepoints
1         A         control         T0
2         A         control         T0
3         A         control         T0
4         B         control         T0
5         B         control         T0
6         B         control         T0
7         A         control         T1
8         A         control         T1
9         A         control         T1
10        B         control         T1
11        B         control         T1
12        B         control         T1
13        A         treated         T1
14        A         treated         T1
15        A         treated         T1
16        B         treated         T1
17        B         treated         T1
18        B         treated         T1
19        A         control         T2
20        A         control         T2
21        A         control         T2
22        B         control         T2
23        B         control         T2
24        B         control         T2
25        A         treated         T2
26        A         treated         T2
27        A         treated         T2
28        B         treated         T2
29        B         treated         T2
30        B         treated         T2
31        A         control         T4
32        A         control         T4
33        A         control         T4
34        B         control         T4
35        B         control         T4
36        B         control         T4
37        A         treated         T4
38        A         treated         T4
39        A         treated         T4
40        B         treated         T4
41        B         treated         T4
42        B         treated         T4

key questions for  DE:

1) time-series DE within a specific group, 2)  time-series DE within specific treatment 3) group vs treatment time series combinations

Different model designs tried:

 ~group+treatment+timepoints+treatment:group

~group+treatment+timepoints+group:timepoints

~group+group:treatment+group:timepoints -  test="LRT", reduced = ~group+group:timepoints

I am having a hard time to understand what design should I use for a complete analysis (I should add in: newbie in R, so learning as I go). 

> sessionInfo()

R version 3.2.2 (2015-08-14)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 7 x64 (build 7601) Service Pack 1

locale:
[1] LC_COLLATE=English_New Zealand.1252  LC_CTYPE=English_New Zealand.1252   
[3] LC_MONETARY=English_New Zealand.1252 LC_NUMERIC=C                        
[5] LC_TIME=English_New Zealand.1252    

attached base packages:
[1] parallel  stats4    stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
 [1] ggplot2_2.2.0              DESeq2_1.10.1              RcppArmadillo_0.7.500.0.0 
 [4] Rcpp_0.12.8                SummarizedExperiment_1.0.2 Biobase_2.30.0            
 [7] GenomicRanges_1.22.4       GenomeInfoDb_1.6.3         IRanges_2.4.8             
[10] S4Vectors_0.8.11           BiocGenerics_0.16.1       

loaded via a namespace (and not attached):
 [1] genefilter_1.52.1    locfit_1.5-9.1       splines_3.2.2        lattice_0.20-34     
 [5] colorspace_1.3-1     htmltools_0.3.5      survival_2.40-1      XML_3.98-1.5        
 [9] foreign_0.8-67       DBI_0.5-1            BiocParallel_1.4.3   RColorBrewer_1.1-2  
[13] lambda.r_1.1.9       plyr_1.8.4           stringr_1.1.0        zlibbioc_1.16.0     
[17] munsell_0.4.3        gtable_0.2.0         futile.logger_1.4.3  memoise_1.0.0       
[21] latticeExtra_0.6-28  knitr_1.15.1         geneplotter_1.48.0   AnnotationDbi_1.32.3
[25] htmlTable_1.7        acepack_1.4.1        xtable_1.8-2         openssl_0.9.5       
[29] scales_0.4.1         base64_2.0           Hmisc_4.0-1          annotate_1.48.0     
[33] XVector_0.10.0       gridExtra_2.2.1      digest_0.6.10        stringi_1.1.2       
[37] grid_3.2.2           tools_3.2.2          magrittr_1.5         lazyeval_0.2.0      
[41] tibble_1.2           RSQLite_1.1          Formula_1.2-1        cluster_2.0.5       
[45] futile.options_1.0.0 Matrix_1.2-7.1       data.table_1.10.0    assertthat_0.1      
[49] rpart_4.1-10         nnet_7.3-12         

 Thanks everyone for your time!!!

hi Sukhi,

You can take a look at the time series example in the workflow:

http://www.bioconductor.org/help/workflows/rnaseqGene/#time

Supposing you didn't have the "group" variable, a difference from that time series example and your data is that you have no treated samples at time 0. So you cannot estimate a "treatment" effect (this is the difference at time 0 between treated and control). Instead you can create a model matrix with just the interaction terms but no "treatment" term:

mm <- model.matrix(~timepoint + treatment:timepoint)

Then you will have to remove the column of 0's from 'mm' because there are no treated samples at time=0. Let's call that cleaned up matrix 'mm.full'. Then, the test for differences in the treated samples across time would be to subset the interaction terms out of 'mm.full', and provide the smaller matrix to the 'reduced' argument. If we call that smaller matrix 'mm.reduced' it would look like:

dds <- DESeq(dds, full=mm.full, reduced=mm.reduced, test="LRT")
res <- results(dds)

Things are complicated by having the "group" variable. In fact there are a couple of possible designs here, and therefore I would recommend you partner with a local statistician who can help you formulate specific designs and contrasts, and interpret results.