A structured matrix factorization framework for large scale calcium imaging data analysis

• DOI: 10.48550/arxiv.1409.2903
• Publication Date: 2014-01-01
• Authors: Eftychios A. Pnev...; Yuanjun Gao; Daniel Soudry; David Pfau; Clay Lacefield; Kira E. Poskanzer; Randy M. Bruno; Rafael Yuste; Liam Paninski
• Keywords: Initialization; Non-negative Matrix Factorization; Synthetic data; Matrix (chemical analysis); Calcium imaging

We present a structured matrix factorization approach to analyzing calcium imaging recordings of large neuronal ensembles. Our goal is simultaneously identify the locations neurons, demix spatially overlapping components, and denoise deconvolve spiking activity each neuron from slow dynamics indicator. The relies on observation that spatiotemporal fluorescence can be expressed as product two matrices: spatial encodes location in optical field temporal characterizes concentration over time. simple for estimating indicator well noise statistics observed data. These parameters are then used set up problem constrained form requires no further parameter tuning. discuss initialization post-processing techniques enhance performance our method, along with efficient largely parallelizable algorithms. apply method {\it vivo} scale multi-neuronal data also demonstrate how similar methods analysis dendritic