NeuroLoom: Modeling Cortical Microcircuits with Spiking Neural Networks

A key problem in neuroscience is understanding how cognitive function emerges from the dynamics of neural circuits. While classical Spiking Neural Networks (SNNs) provide a viable, biologically-based template for modeling these dynamics, their value can emerge only through comparisons to experimental data. Here, we present NeuroLoom, an open-source, customizable framework to design, validate, and analyze models of cortical microcircuits against experimental data. NeuroLoom provides a complete end-to-end workflow that incorporates a model library of standard neurons and synapses and programmatic access to the Allen Brain Observatory experimental recordings through the AllenSDK, making it simple to construct, fit, and analyze model microcircuits. We demonstrate the capabilities of the framework through a pilot experiment in which a cortical microcircuit model of adaptive exponential integrate-and-fire (AdEx) neurons is constrained and validated against in-vivo electrophysiology recordings from the mouse visual cortex. The framework additionally contains runnable examples of important cortical phenomena, including models of synaptic plasticity, cognitive performance, and neuromodulation, making NeuroLoom a step towards building stronger, verifiable models of the brain, and is an extensible, interpretable framework to facilitate advancing interdisciplinary research.