A Weighting Profile Method for Protein-RNA Interaction Prediction

A typical feedforward network has units arranged in a distinct layered topology. Units are connected to one another and each connection is associated with a real number, which is called the weight of the connection. During network training, the connection weights are adjusted in order to correctly classify the training data. The network weights are basically dependent on the training data set and give evidence of which inputs were more influential in the network. That is, the larger the magnitude of a weight is, the higher its level of participation in the solution is. If the value of a weight is near zero, the connected input unit may be not important to the solution. In this study, we propose a method to analyze and qualify a large set of the network weights which are trained on protein sequence profiles, and explore the feasibility of utilizing the qualified information to improve the prediction performance for protein-RNA interaction.