[BioC] LDA and qPCR data

[David martin]

Hi,
I have generated some data using rt-pct;
 From 50 samples (divided into three groups healthy,chronic,disease) i 
have tested all samples for th expression of 50 genes. I want to find 
the most suitable combination of genes that will help to discriminate 
three groups healthy, chronic and disease.

For that i have started by running an anova, followed by TukeyHSD test 
and have identified  gene1 to 10 as differentially expressed in group 
disease from both chronic and healthy. I have evaluated the ROC plots 
for each of these genes and noticed that a three of those can help to 
better discriminate group chronic from disease. So the conclusion would 
be that gene1+gene2+gene3 are my best markers between group chronic and 
disease.

Everything is ok until here but i want to statistically show that three 
genes are the best combination markers. Should i run individual LDA 
(between pair genes) or is there any multi-class LDA test that i could 
use to show that this is the best combination of markers to 
differentiate chronic from disease in ly set of 10 genes ?

thanks for your help and tips
david