[BioC] cDNA microarray Questions
naomi at stat.psu.edu
Tue Mar 2 05:26:59 MET 2004
Many people delete low quality genes based on either comparison between
background and foreground, comparison between the mean and median
foreground, or visual inspection. I am reluctant to put a guess as to a
good value for correlation based on the data I have worked with, which have
been low-quality pilot studies with many spots deleted due to quality.
However, it surely depends on the biological system - inbred lines in a
highly controlled environment should be less variable, outbred and natural
populations much more. Also, the correlations that you see in the
literature are generally after normalization - global normalization will
not affect the correlation, but lowess (or loess) certainly will. 50%
seems low compared to what I have heard - but then, I have mostly heard
about model organisms under controlled conditions.
At 10:12 AM 3/1/2004, michael watson (IAH-C) wrote:
>As low as 0.5 for pearson and 0.8 for spearman
>But I'm not holding up my data as a shining example
>Surely if there is a lot of natural variation in the biological system,
>then you're going to get large variation between biological replicates.
>From: Naomi Altman [mailto:naomi at stat.psu.edu]
>Sent: 28 February 2004 05:20
>To: michael watson (IAH-C); 'rwin qian'; bioconductor at stat.math.ethz.ch
>Subject: RE: [BioC] cDNA microarray Questions
>What do you mean by "very low"?
>At 04:24 AM 2/27/2004, michael watson (IAH-C) wrote:
> > >The correlation coefficients are very low. Is that the normal case?
> > >Do I need to delete some poor quality genes before any analysis and
> > >what rule should I use?
> >Which correlation coefficient are you using? I regularly see very low
> >pearson correlation coefficients between biological replicates but that
> >can be put down to natural biological variation - no two organisms are the
> >same, right? BUT if you start seeing low correlations between technical
> >replicates (e.g. replicate samples from the same
> >animal/tissue/organ/whatever) then that indicates that you have a lot of
> >variation in your technology, which is bad.
> >ON another note, I always find the Spearman Rank Correlation Coefficients
> >to be much higher.
> >Bioconductor mailing list
> >Bioconductor at stat.math.ethz.ch
>Naomi S. Altman 814-865-3791 (voice)
>Bioinformatics Consulting Center
>Dept. of Statistics 814-863-7114 (fax)
>Penn State University 814-865-1348 (Statistics)
>University Park, PA 16802-2111
>Bioconductor mailing list
>Bioconductor at stat.math.ethz.ch
Naomi S. Altman 814-865-3791 (voice)
Bioinformatics Consulting Center
Dept. of Statistics 814-863-7114 (fax)
Penn State University 814-865-1348 (Statistics)
University Park, PA 16802-2111
More information about the Bioconductor