[BioC] ebayes and identifying outlyingly large variances

[Ramon Diaz-Uriarte]

Dear All,

For a data set I am working on, besides ordering coefficients by "differential 
expression", we are interested in getting a quick idea of whether the 
variance of some genes is unexpectedly large.

I thought about applying the expression in p.10 of the tech. report version of 
Smyth's "Linear Models and Empirical Bayes Methods (...)" 
(http://www.statsci.org/smyth/pubs/ebayes.pdf),

where  s^2 ~ s_0^2 F_{d,d0}

For each gene, we can compare the error variance from the linear model with 
the corresponding distribution from the above expression.

(in R, if lmFit in fit and ebayes in fit.eb:

p.vars <- pf(fit.eb$s2.post/fit.eb$s2.prior, fit$df.residual, fit.eb$df.prior)

)

Does this make sense?

R.




-- 
Ramón Díaz-Uriarte
Bioinformatics Unit
Centro Nacional de Investigaciones Oncológicas (CNIO)
(Spanish National Cancer Center)
Melchor Fernández Almagro, 3
28029 Madrid (Spain)
Fax: +-34-91-224-6972
Phone: +-34-91-224-6900

http://bioinfo.cnio.es/~rdiaz
PGP KeyID: 0xE89B3462
(http://bioinfo.cnio.es/~rdiaz/0xE89B3462.asc)