Mo Shakiba

Abstract

“Understanding how neural circuits drive sensory processing and decision-making is a central neuroscience challenge. Traditional Recurrent Neural Networks (RNNs) capture temporal dynamics but fail to represent the structured synaptic architecture seen in biological systems. Recent spatially embedded RNNs (seRNNs) add spatial constraints for better biological relevance, yet they do not fully exploit detailed anatomical and functional data to enhance task performance and neural alignment. We introduce a bio-inspired RNN that integrates detailed anatomical connectivity and two-photon calcium imaging data from the MICrONS dataset. Using neuronal positions, synaptic connections, and functional correlations we constrain our model with biologically informed weight initialization, and communicability calculations to promote realistic network properties. Our findings demonstrate that incorporating biological constraints into RNNs significantly boosts both task performance and the emergence of realistic network properties, mirroring actual neural architectures.”