How Does Detection Of Differentially Expressed Genes In Time-Course Microarray Experiments Work?

Question

[Student]

When I asked people about differentially expression of genes in time-series experiments I got the answer: use limma.

But how/why does this work in general? Let's say we have the time-course 0h 3h 6h 1day 1week 2weeks. How can we throw differences like 0h - 3h and 1week - 2weeks in one linear model? What information does 't' carry, that helps finding differentially expressed genes in 't+1' and 't+4', respectively?

What method would you use for e.g. 0h 3h 6h 1day 1week 2weeks and e.g. 0h, 3h, 24h, 48h?

Thank you very much.

Answer 1

[Michi]

To know if two genes are differentially expressed, one has to make a comparison between a "normal" state and a mutation (or anormal) state. Comparing these two states at different time points, you can extract which genes behave in a different manner in the states by simple comparisons (e.g. ratio).

So it may be that a certain group of genes is always much more transcribed in the anormal state - this differentially expressed genes are called upregulated

Look the limma guide on page 50: Time-course experiment: They compare a mutant against a wild-type phenotype. In cancer it is normally called: normal vs cancer phenotype comparison, etc..

I hope you are a bit helped ;-)

[Student]

Thanks for the answer, Michi.

I do get the differentially expression part, but still not how the time variable helps us in a reliable manner.

As far as I know, there is not a 'turning on' of e.g. cancer genes. E.g. a differentiation process goes through a daily, sometimes weekly process, turning on and off different genes all the time. I do not think that there is a linear phase from 'normal' to 'cancer', or is there? because I think the model is treating the problem like this

[Michi]

You are right and right. If a gene is turned on and off in cell development and differentiation - or even better they are expressed in cyclic manners. The time variable is of value in a very well defined model and the experiment has to be well synchronized - so you can expect a certain behaviour of the genes you are watching.. maybe this lab can give you an idea how it can be useful: http://www.bahlerlab.info/projects/cellcycle/

[Ali]

limma although is very powerful in normalization and identification of differential expression, I don't recommend it for time course analysis at all.

[Zhilong Jia]

@Ali, Could you show the reason? Thanks.

Answer 2

[boczniak767]

Student, I advise you to use limma for normalization and maSigPro package for time-course analysis. It involves regression analysis and works fine (except clustering).

It also draws useful plots comparing analyzed profiles.

[Student]

thanks for that, Maciej