test p values less than 0.01. When the variance test p value is less

, it means that the null hypothesis of identical variance between

ers of expressions of a bimodal gene is rejected. In total, 27%

bimodal genes had significantly heterogeneous variances across

ers of expressions. However, this will not happen when using the

misation test approach. The gap maximisation test does not need

tribution assumption.

The distribution of the variance test p values for the identified bimodal genes

imodal Index test for the letrozole drug data.

ther problem of the bimodal index test is the uncertainty. This is

it involves a clustering process employing the mixture model

m, which can lead to different clustering results in different runs.

was applied to the breast cancer letrozole drug treatment data ten

n sets of predictions had different numbers of discovered bimodal

gure 6.40 shows the result. It can be seen that the minimum

f the discovered bimodal genes in one model was 8,528 and the

m number of the discovered bimodal genes in the other model was

The difference was 1,903, which was 18.24% of the maximum

of the discovered bimodal genes. This data set has 58 replicates,

not small. Many gene expression data would have a much smaller

of replicates. Therefore such an uncertainty may be severely

On the other hand, the gap maximisation test algorithm has no

onal assumption, and the test result will be almost certain in

runs.