Neural Network Embedding of Functional Microconnectome.

Abstract

Our brain works as a vast and complex network system. We need to compress the networks to extract simple principles of network patterns and interpret these paradigms to better comprehend their complexities. This study treats this simplification process using a two-step analysis of topological patterns of functional connectivities that were produced from electrical activities of ~1000 neurons from acute slices of mouse brains [Kajiwara et al. 2021] As the first step, we trained an artificial neural network system called neural network embedding (NNE) and automatically compressed the functional connectivities. As the second step, we widely compared the compressed features with 15 representative network metrics, having clear interpretations, including not only common metrics, such as centralities clusters and modules but also newly developed network metrics. The result demonstrates not only the fact that the newly developed network metrics could complementarily explain the features of what was compressed by the NNE method but was previously relatively hard to explain using common metrics such as hubs, clusters and communities. This NNE method surpasses the limitations of commonly used human-made metrics but also provides the possibility that recognizing our own limitations drives us to extend interpretable targets by developing new network metrics.