MINING PROTEIN REGULATORY RELATIONSHIPS USING NEURAL NETWORK METHODS FOR EARLY PREDICTION OF SARS

This paper proposes to model protein regulation networks associated with severe acute respiratory syndrome (SARS) for early prediction of SARS. In the approach, specific to a patient group, a regulatory network is simulated using a fully-connected neural network and is optimized towards minimizing a novel energy function that is defined as a measure of disagreement between the input and output of the network. The nonlinear version of the network is achieved by applying a sigmoid function. Experimental results show that the proposed approaches can capture regulatory patterns associated with SARS and efficiently implement early prediction of SARS.