Cuffdiff time series

[bsuac6]

Hi,

I am using RNA-seq data. I have 3 time points in my data 6 months (young), 12 months (middle) and 28 months (old). I want to do a differential expression analysis across the 3 time points. I have used cuffdiff to do this:

cuffdiff -o /time -L young,middle,old -b /RGSC3.4.61.dna.toplevel.2.fa -u -N -T --library-type fr-secondstrand /allrat.combined.gtf /1.sam,/2.sam,/3.sam
/4.sam,/5.sam,/6.sam /7.sam,/8.sam,/9.sam

The output I get is a comparisons between "young and middle" and "middle and old". Does this seem correct? I would have expected a comparison of "young and old" whilst taking into account the differences in middle?

Thanks

[dahlo]

Hi,
The output I get is a comparisons between "young and middle" and "middle and old". Does this seem correct? I would have expected a comparison of "young and old" whilst taking into account the differences in middle?

Thanks

Hello there

I am pretty sure cufflinks can only do pairwise comparisons. If you enable the time series mode (-T/--time-series), it will only pairwise compare each time point with the next.

An analysis across all time points will have to be done with another program

[bsuac6]

Thanks for your reply. I suspected this was the case. I have used edgeR GLM to do this analysis as well but it doesn't give me the contrasts I would like e.g. linear regression of change with increasing age which takes into account the expression at middle age. I am told this is a function of the maths because I only have three time points. Suppose pairwise comparisons it is!

[rhcr56]

comparing a time series with pairwise comparisons

Just a word of caution about comparing a time series via pairwise comparisons. Sometimes the results can be misleading. For example, a gene whose RNA abundance linearly increases/decreases over time may be significantly different from the first time point to the last, yet is not significant in any of the sequential pairwise comparisons. With increasing numbers of time points, the results can get even more convoluted! If you only have three time points and were still interested, I suggest leaving the -T option out of the command line because Cuffdiff will now do all pairwise comparisons "young vs middle", "middle vs old", and "young vs old". Hope this was helpful. If you have any questions let me know.

[bsuac6]

Thanks for your comments. In fact, I did end up doing my analysis as you suggested, it worked quite well - though I am glad I don't have more than three time points!

[dvelayutham]

Hello bsuac6

After leaving the -T command suggested by 'rhcr56', you would have probably end up with "young vs middle", "middle vs old", and "young vs old" comaparisions from the single command.

But have you ever tried doing one pairwise comparision at a time, So three separate cuffdiff commands to produce three different comparision ?? Did the no. of sig genes increase ??

I have my experient like you with three-four time points, and I wonder which is the ''Best'' option
Any suggestion is appreciated.

[anthurium]

Hi,
I also have some samples to analyses in different combinations... I would like to understand more clearly how the significance test works when you do single comparisons (one comparison= one cuffdiff command) or when you do multiple comparisons in the same command

with my data, when I run multiple comparisons all together in the end I get less significant genes than doing it one at once

can anyone explain how the calculations are made?

[Kseniya]

a very good question, I am also interested in this. With my data I noticed the same behavior. When I decrease the number of samples, I get different fpkm and consequently p-value and number of differentially expressed genes.