Rapid, interpretable, data-driven models of neural dynamics using Recurrent Mechanistic Models

Obtaining predictive models of a neural system is notoriously challenging. Detailed models suffer from excess model complexity and are difficult to fit efficiently. Simplified models must negotiate a tradeoff between tractability, predictive power and ease of interpretation. We present a new modelling paradigm for estimating predictive mechanistic-like models of neurons and small circuits that navigates these issues using methods from systems theory. The key insight is that membrane currents can be modelled using two scalable system components optimized for learning: linear state space models, and nonlinear artificial neural networks (ANNs). Combining these components, we construct two types of membrane currents: lumped currents, which are more flexible, and data-driven conductance-based currents, which are more interpretable. The resulting class of models — which we call Recurrent Mechanistic Models (RMMs) — can be trained in a matter of seconds to minutes on intracellular recordings during an electrophysiology experiment, representing a step change in performance over previous approaches. As a proof-of-principle, we use RMMs to learn the dynamics of two groups of neurons, and their synaptic connections, in the Stomatogastric Ganglion (STG), a well-known central pattern generator. We show that RMMs are efficiently trained using teacher forcing and multiple-shooting. Due to their reliability, efficiency and interpretability, RMMs enable qualitatively new kinds of experiments using predictive models in closed-loop neurophysiology and online estimation of neural properties in living preparations.