[BioC] array CGH

[Sean Davis]

On Friday 13 October 2006 10:04, Lisa Luo wrote:
> Thank you, Sean and all others who replied to me.
>
>   First for segmentation method, I am not satisfied with DNAcopy.  A
> extreme in this case is that when I have a break point near the middle, it
> could not detect it if I used the SD trim method.  Or it gave too many
> breaking points.  

You may have to change parameters multiple times to get the results that you 
like.  Also, you might consider using a merging function like that 
implemented in snapCGH rather than using trimming, but others may not agree 
here.

> Also, it can not detect full chromosome change that you 
> have to rely on the ratio.

The goal of the segmentation methods is not to call gains and losses, but to 
determine regions of constant copy number.  There is another step in the 
process to determine whether a copy number estimate actually represents a 
gain or loss.

>   Is BAC array CGH single strand PCR arrays?  If so, should repeat affact
> the result?  In my tumor samples with opposite sex for control, I found
> that the ratio for X varied from sample to sample.  In one sample, it is
> even 1.  The explaination I got is that chrX has more repeat regions.  Are
> you suggesting that DNA quality may play a role in this?

Of course quality plays a role in the array result--both quality of the arrays 
and quality of the DNA.  If you have arrays on which you have a male and a 
female hybed and the mean of the X-chromosome is the same of the mean of the 
autosomes, there are a few explanations:

1)  Sample mixup--the samples are actually the same sex
2)  One of the samples has either gained/lost an X chromosome (cell line 
artifact, if the samples are from cell lines)
3)  The processing/hybing of the array failed for some reason
4)  Probes are somehow mismapped, either by you or by the person supplying the 
annotation.
5)  There actually IS a difference between the X-chromosome and the autosomes, 
but the difference is small or difficult to notice given the noise; for 
practical purposes, the X-chromosome/autosome ratio represents signal and 
noise can be quantified in numerous ways (SD, MAD, etc).  Arrays with low 
signal-noise-ratio for whatever reason will look like there is very little 
difference between the autosomes/X-chromosome.

>   I am planning to look at each BAC at one time, comparing tumor vs normal.
>  Is this a good idea?

This sounds fine, but I would use the data calculated from the segmentation 
results as the input to your test; this will reduce the noise associated with 
individual probes.  Also, note that the data are not likely to be normally 
distributed at a given probe across samples, since each sample is potentially 
drawn from a different population (mean copy number).

Sean