Gene network modelling using computational method by integrating with prior knowledge

One of the aims of system biology is to infer gene networks that represent interaction between genes from biological data. Many computational methods have been developed to infer gene networks using microarray data in order to understand cellular processes and relations between genes. Gene network inference will generate hypothesis about novel gene functions and also verify known gene functions. However, network inference task is challenging due to the exponential increase of the search space as more variables are used for inference. This task was originally performed using gene expression profiles from microarray as the single input. The accuracy of inference results depends on the careful selection of the input variables. This paper proposed the use of prior biological knowledge and rough sets attribute reduction to select the input variables for gene network inference. Firstly, Self-Organizing Maps (SOM) is used to cluster the microarray data. Feature selection will be employed in clustering analysis, by eliminating the least interesting and highlight the most interesting features. Rough set theory incorporated with prior knowledge to model is applied to the top ranked features prior to gene network inference. This proposed method is expected to infer reliable gene networks with higher prediction accuracy using a small number of features.