Even among the subsets that are identified as important, the identification of a few targets can be difficult and subjective. Another major drawback of these approaches is the tedious manual preparation and updating of the data sets. While data sets are available for Affymetrix chip for some species, for other platforms, such as the custom cDNA microarrays, one would need to manually define the gene sets, which can delay procurement of downstream information or introduce errors. Therefore, such techniques are useful when the expressions of many genes of an important, causative pathway change. For other Presapogenin-CP4 situations where there is change of only a few, rate-controlling genes, such approaches may not be as informative. Nevertheless, applications of these techniques have lent support to the notion that incorporation of prior knowledge could either improve the classification efficiency or identify more relevant biological processes. We demonstrated the proposed method by applying it to identify the genes that are likely involved in the toxicity of FFAs, in particular saturated, palmitate, and TNF-a. Our experimental results showed that our proposed method is able to identify the group of toxic experimental conditions, and identify the genes that are relevant to the toxic conditions. The main idea behind the Epimedin-A1 mixture models is to cluster the experimental conditions into an optimal number of subgroups and build a different regression model that relates the gene expression data to a cell response for each subgroup. The clustering of experimental conditions, however, is based on their regression weights. For example, two experimental conditions will be grouped into the same cluster if they share similar regression weights. However, the regression weights of each experimental condition would also depend on the clustering results because a regression model can be built only for a group of experimental conditions. Hence, the technical challenge of regression mixture model lies in resolving this dilemma. We applied Expectation Maximization algorithm to effectively resolve this problem.