Hello community,
I have a mice RNASeq time course experiment with RNASeq samples at time points day0,day2,day4,day6.
On day0 no treatement(lps) was given to control group(4 mice) and target celltype from all 4 mice was sequenced (bulk RNASeq).
3 groups were subjected to treatment and same target celltype from 3 different groups at different time points (day 2, day 4 and day 6 respectively) was sequenced.
I want to find how the mice respond to lps over time, I donot have controls for everytime point.
design
sample condition day treat
WT_D0B_1 D0B_nolps D0B nolps
WT_D0B_2 D0B_nolps D0B nolps
WT_D0B_3 D0B_nolps D0B nolps
WT_D0B_4 D0B_nolps D0B nolps
WT_D2A_1 D2A_lps D2A lps
WT_D2A_2 D2A_lps D2A lps
WT_D2A_3 D2A_lps D2A lps
WT_D2A_4 D2A_lps D2A lps
WT_D4A_1 D4A_lps D4A lps
WT_D4A_2 D4A_lps D4A lps
WT_D4A_3 D4A_lps D4A lps
WT_D4A_4 D4A_lps D4A lps
WT_D6A_1 D6A_lps D6A lps
WT_D6A_2 D6A_lps D6A lps
WT_D6A_3 D6A_lps D6A lps
WT_D6A_4 D6A_lps D6A lps
I haved used combined format of day and treatment as condition and used it to model the experiemtnal design as below
dds <- DESeqDataSetFromMatrix(countData = countsdata,
colData = conditions,
design= ~ condition)
# LRT test
dds_lrt <- DESeq(dds, test="LRT", reduced=~1)
res_LRT <- results(dds_lrt)
#
# to check all the deg's across times
d2_lps_vs_d0_nolps<-results(dds_lrt, contrast=c("condition", "D2A_lps", "D0B_nolps"),alpha = 0.05)
d4_lps_vs_d0_nolps<-results(dds_lrt, contrast=c("condition", "D4A_lps", "D0B_nolps"),alpha = 0.05)
d6_lps_vs_d0_nolps<-results(dds_lrt, contrast=c("condition", "D6A_lps", "D0B_nolps"),alpha = 0.05)
There is almost 99% overlap among the DEG analysis across all groups LRT test and other individual contrasts?
Do I have my experimental design marix correct for model fitting?
Yes Mike, here is the link
https://ibb.co/fTVEVK
It showed that major variance was due to LPS, however, the samples are also separated among the time points.
And what exactly is the overlap between the four sets of DEG? I would expect that perhaps day 4 vs day 0 and day 6 vs day 0 may be similar and also similar to the LRT as this is the main axis of variation.
Its strikingly similar , shown below
https://ibb.co/c7Anje
Oh I just noticed, you are not setting test="Wald". At the top of your results tables it will show that you have the same LRT p-values each time (the only difference being that you changed the target alpha for filtering).
Specifying test="Wald" is necessary if you want to switch from test="LRT", which is set when you run DESeq(). This is described in the man page for results() under the 'test' argument.
Thank you Mike,
One last question for time series analysis, what are the disadvantages if I do not have a matched time point control and how can I address it if I won't be able to redo the experiment.
This is hard to state in general.
Often it means assuming that there is no difference between the groups at time 0. Having treated samples at time 0 helps to correct if there are systematic differences between the series, for example if the treated samples share an environment that differs from the untreated.
Thank you Mike !!