Inferring state-dependent functional circuit motifs using higher-order interactions analysis

Abstract

Analysing higher-order interactions among simultaneously recorded neurons can provide crucial insights into neural network dynamics. Recent technological advances have enabled large-scale, long-term neuronal recordings, but analysis of such datasets often relies on simpler statistics due to computational and statistical challenges in assessing higher-order interactions. Here, we developed CHOIR, an efficient and reliable method for calculating higher-order interactions from large-scale neuronal recordings. We then used the inferred HOIs to uncover the underlying functional connectivity, differentiating between connectivity motifs in the space of pairwise and triplet-wise interactions. We found that this approach could successfully distinguish stationary and running states, sleep and awake states, and neuronal ensembles with distinct activity patterns in mice. Furthermore, we identified potential circuit architectures underlying different higher-order interactions, which we confirmed through simulations of large-scale spiking networks with specific subnetwork connectivity. Applying CHOIR to a causal manipulation dataset further confirmed the role of lateral inhibition, a key inhibitory motif, in generating specific HOI patterns. Our work provides a systematic analysis of higher-order interactions in diverse datasets and suggests that HOIs can reveal circuit motifs underlying neural dynamics across brain areas and brain states.