Hi,

I am using the fourCseq package to analyse 4C data, and  I have several
questions:

• in the vignette you do precise that if too many fragments have no reads, this can result in a shift of the zscore distribution in negative value, making the distribution no more centred around 0, and therefore to a errors in the calculation of z-scores. If this is the cse, then the data are just too sparse to allow a proper calculation of the zscore and cannot be used? or is there anyway to correct it?
• is it possible to identify interraction in trans, ie on another chromosome? I could run the analysis on all the genome by specifyign all the chromosomes on the "referenceGenomeFile" arguments, but isn't does it biais the zscores calculation? (because most of the trans fragments should have no reads). Or is the fit used for the zscore calculation done only on the automatically/manually defined region around the view point when calling the function getZscores() ?

Many thanks in advance,

-- 

Samuel COLLOMBET 

PhD student 

Computationnal systems biology Lab (Pr. Thieffry)

Institut of Biology at the Ecole Normale Superieure

Hi Samuel,

concerning your first question you might have to increase the threshold that you use to filter out fragments in the first place. If you are left with a reasonable number of fragments you can continue with the analysis.

There is no implementation to identify trans interactions in this package. And the zscore calculation works only for the viewpoint chromosome.

Best regards,

Felix

Hi,

In some cases I see that adjacent fragment are variable in two conditions i.e. one is high in one condition and zero/low in the other and directly next to it it is the opposite. This gives me a lot of differential interactions, however the peaks are indifferent. I don´t think so this is due to biology because these fragments are too close belonging to the same promoter etc..

What I tried that after counting at all fragments I merged 3,5,... fragments, sum the read counts and test for difference. I lose spatial resolution and if I do it too much I average out the peaks but I get rid of the noise.

Do you think it is a good solution for such samples?

How should I determine how many frags to merge (noisy interactions gone, peaks are still there)?

Cheers,

Tamas

Dear Tamas

such a smoothing (sums over local windows) is reasonable to stabilize the data; the choice of the window size will depend on your genome, cutting enzyme(s), coverage etc., I am not sure there is an objective algorithm for its choice, but you could tune it from know positive and negative controls.

Best wishes

Wolfgang