Localized semi-nonnegative matrix factorization (LocaNMF) of widefield calcium imaging data

Abstract

<jats:p>Widefield calcium imaging enables recording of large-scale neural activity across the mouse dorsal cortex. In order to examine the relationship of these neural signals to the resulting behavior, it is critical to demix the recordings into meaningful spatial and temporal components that can be mapped onto well-defined brain regions. However, no current tools satisfactorily extract the activity of the different brain regions in individual mice in a data-driven manner, while taking into account mouse-specific and preparation-specific differences. Here, we introduce Localized semi-Nonnegative Matrix Factorization (LocaNMF), a method that efficiently decomposes widefield video data and allows us to directly compare activity across multiple mice by outputting mouse-specific localized functional regions that are significantly more interpretable than more traditional decomposition techniques. Moreover, it provides a natural subspace to directly compare correlation maps and neural dynamics across different behaviors, mice, and experimental conditions, and enables identification of task- and movement-related brain regions.</jats:p>

Distributed context-dependent choice information in mouse dorsal-parietal cortex

“…When seeding on a given area, locaNMF decomposes the signal into a sum of separable spatial-temporal tensors, with spatial components constrained by the seeding region and temporal components representing the scaling amplitudes of the spatial components. These temporal vectors are potentially more informative than a single vector computed as the average across spatial locations (pixels) within a given area 21 . We aligned all imaging sessions according to the Allen Common Coordinates Framework (CCF 29 , Fig.…”

“…As the animals engaged in a complex variant of a two-alternative forced choice (2AFC) orientation discrimination task 20 , we performed mesoscale imaging of neural responses (GCaMP) from the mouse posterior cortex. Using recent tensor decomposition methods 21 and activitymode analysis 22 we isolated choice signals from concurrently represented sensory and motor-related activations, and characterized their distinct spatial and temporal properties across cortical areas. In a reduced space of multi-area activations, choice signals defined an embedding subspace for left or right (L/R) decisions that was near-orthogonal to that of sensory signals and movement components and was modulated by task difficulty.…”

Neuroscience Cloud Analysis As a Service

“…cost of ownership (TCO) metric (Morey and Nambiar, 2009) that includes the purchase cost of local hardware, plus reasonable maintenance costs; see Methods for full details. Figure 4 presents time and cost benchmark results on four modern data analysis algorithms hosted on NeuroCAAS: CaImAn, a toolbox for analysis of calcium imaging data (Giovannucci et al, 2019); DeepLabCut (DLC), a method for markerless tracking of pose from behavioral video data (Mathis et al, 2018); Penalized Matrix Decomposition (PMD), a method for denoising and compressing functional imaging data (Buchanan et al, 2018); and Localized Non-negative Matrix Factorization (LocaNMF), a method for demixing widefield calcium imaging data (Saxena et al, 2020). Each analysis presented in Figure 4 highlights a different strength of NeuroCAAS compared to local infrastructure (see Methods for implementation details).…”

Extracting neural signals from semi-immobilized animals with deformable non-negative matrix factorization

“…Similar to the approach taken by [Saxena et al, 2019], we can introduce a regularizer to encourage spatial proximity of non-parametric components to a set of predefined centers that is defined by the statistical atlas of neuron positions. Given a (non-negative) constraint matrix D ∈ R d×k such that each column encodes the allowable occupancy maps of each of the k cells, the regularizer is:…”