Exploring Protein Regulations with Regulatory Networks for Cancer Classification

This paper proposes a novel modeling technique for understanding cancer signal pathway and applies to cancer classification. In the approach, specific to a cancer group, a regulatory network is constructed between biomarkers and is optimized towards minimizing its energy function that is defined as disagreement between input and output of the network. The non-linear version of this network is achieved by imposing a sigmoid kernel function. The proposed approach is tested on protein profiling data of nasopharyngeal carcinoma and is compared with support vector machines with linear and radial basis function kernels.