Investigating Feedback Loops in Synthetic Neural Networks by a Nonparametric Identification Method

Abstract

Feedback circuits are crucial dynamic motifs which occur in many intra-cellular and inter-cellular regulatory networks. In this paper, an effective nonparametric identification method, Non-causal Impulse Response Component Method (NIRCM) is developed to identify feedback loops embedded in biological neural networks, which uses only time-series experimental data. The NIRCM, based on correlation identification and spectral factor analysis, provides a non-causal component criterion for the identification of feedback loops. Significant non-causal components of the impulse response sequences observed in the negative time axis imply an existence of feedback loop. The proposed identification method was applied to several 2-node SRM (Spike Response Model) networks. For these synthetic models, NIRCM correctly implies the existence of feedback loops and shows their effectiveness of feedback loop identifications.