ulation of data through maximising the difference between two

of an empirical distribution [Hartigan, 1985; Hartigan and

1985]. Its null hypothesis is the unimodality.

udy the subpopulation formation problem in biomedical practices,

xpression bimodality pattern discovery project can be based on a

ject approach or a dual-object approach. The dual-object

es address the cooperation between two genes such as an

e and a suppression gene in relationship to a disease development

, et al., 2016]. The single-object approaches take one gene for the

his chapter mainly focuses on the single-object approaches for

pression bimodality pattern discovery. Most single-object

es assume that the expressions of a gene follow a mixture of two

ons [Hellwig, et al., 2010].

measure used by the PACK algorithm is similar to the Kurtosis

chendroff, et al., 2006], where the bimodality of a gene is tested

tosis profile, which is either negative or positive. A positive

stands for a super-Gaussian and a negative measure represents a

sian. The bimodality index approach, on the other hand, assumes

xpressions of a bimodal gene follow a mixture of two Gaussians

t al., 2009]. The algorithm was enhanced by the Markov chain

arlo simulation. The relative discrepancy measure is also based

sumption of a Gaussian mixture [Bessarabova, et al., 2010]. By

ng a ratio between a bimodal distribution and a unimodal

on, the likelihood ratio between them is used for the bimodality

iscovery [Hu, 2008]. In addition to the use of a measure, cluster

has also been used to discover a bimodal distribution [Gormley

ren, 2008].

e likelihood ratio test approach

ihood ratio test (LR) assumes that gene expressions follow a

distribution and the bimodality only occurs in the case

n [Hu, 2008]. Suppose a gene expression vector is denoted by z.

nt vector of a control expression vector x and a case expression