Decoding Neuronal Ensembles from Spatially-Referenced Calcium Traces: A Bayesian Semiparametric Approach

TL;DR

A fully Bayesian, semiparametric model that jointly infers spiking activity and identifies functionally coherent neuronal ensembles from calcium traces is introduced that uncovers spatially structured co-activation patterns among neurons and can be employed to reveal how ensemble structures vary with the animal's position.

Abstract

Understanding how neurons coordinate their activity is a fundamental question in neuroscience, with implications for learning, memory, and neurological disorders. Calcium imaging has emerged as a powerful method to observe large-scale neuronal activity in freely moving animals, providing time-resolved recordings of hundreds of neurons. However, fluorescence signals are noisy and only indirectly reflect underlying spikes of neuronal activity, complicating the extraction of reliable patterns of neuronal coordination. We introduce a fully Bayesian, semiparametric model that jointly infers spiking activity and identifies functionally coherent neuronal ensembles from calcium traces. Our approach models each neuron's spiking probability through a latent Gaussian process and encourages anatomically coherent clustering using a location-dependent stick-breaking prior. A spike-and-slab Dirichlet process captures heterogeneity in spike amplitudes while filtering out negligible events. We consider calcium imaging data from the hippocampal CA1 region of a mouse as it navigates a circular arena, a setting critical for understanding spatial memory and neuronal representation of environments. Our model uncovers spatially structured co-activation patterns among neurons and can be employed to reveal how ensemble structures vary with the animal's position.

Figures (20)

• Figure 1: Left panel: mouse movements within the environment (continuous line). The color of the background corresponds to the two regions of the arena. Right panel: example fluorescence traces of 20 representative neurons. The background colors correspond to the mouse position in the two regions over time. The orange highlights correspond to a representative window.
• Figure 2: Estimated spike amplitudes over time for active neurons. The traces are sorted according to the neuron's cluster allocation (Clusters 2 to 10, left labels).
• Figure 3: Left, center, and top-right panels: observed calcium traces sorted (and colored) by estimated cluster allocation (only active neurons). Points correspond to the times of the detected firing events. Bottom-right panel: neurons' location in the hippocampus, colored according to the estimated cluster allocation.
• Figure 4: Left and center panels: mouse movements within the arena (grey continuous line) and neuronal activity intensity at that location. Right panels: location of the corresponding neurons in the hippocampus: Neuron A (triangle); Neuron B (diamond); Neuron C (square); Neuron D (down triangle).
• Figure 5: Heatmaps showing the spatial distribution of the clustering complexity and variability. Each point of the mouse trajectory is reweighted by the mode (left panel) and variance (right panel) of the posterior distribution of the number of clusters.