Figure 8

Ordered genes and resulting percentage correct classifications. For a given number of genes, a set of genes most able to discriminate between disease states is selected. Simulations of noisy readings of patient gene expression levels are performed for varying levels of noise. Each colored line plots the percentage of correct classifications versus the number of genes used to make the classification for one particular percentage of random values in the simulated readings. With no noise, very few genes are required to discriminate perfectly. With high noise levels (say, 50%), even 1000 genes cannot reliably discriminate well.