A new graph-theoretic approach to neural structure identification with implications for information-theoretic optimization is proposed for dynamic link architecture neural networks. The method addresses the organization and interconnection of feature-representing multicellular units (MCUs) at the network architecture level. It incorporates indices for the identification of both dynamic and static data pathway and binding code links between neurons within an MCU and between different MCUs. It also proposes a new objective function for optimization of a measure of conditional entropies and mutual information properties of the system. The method is based on an extension of a method of graph-theoretic information flow analysis recently proposed by Akuzawa and Ohnishi, here called Akuzawa-Ohnishi Analysis (AOA).