Sparse Granger Causality: Optimization and Application for Connectivity Estimation in Juvenile Zebrafish
Keywords
Loading...
Authors
Issue Date
2021-07-01
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
This work is based on a novel Sparse Granger Causality inference framework,
originally demonstrated to efficiently identify sparse networks of neurons in the primary
auditory cortex of mice involved in decision-making (Francis et al., 2018). Significant
improvement in runtime has been achieved through efficient use of memory resources by
introducing vectorization, optimized parallelization, and weight threshold during connectivity
inference. Depending on the number of neurons and cores available on a multi-threaded
machine, a speedup of > 10-fold was observed with minimal loss of accuracy compared to
the original formulation of this inference framework. Ridge regularization at the level of
cross-history covariate matrices was further introduced for stabilization of the inference.
Next, functional connectivity networks obtained from 2-photon calcium imaging data from
the habenula and telencephalon (~1000 cells) of juvenile zebrafish during task and rest were
thoroughly investigated. Despite the significant number of connections inferred, the number
of connections shared between experimental blocks was observed to be extremely low, while
permutation testing indicates that persistent connectivity is highly unlikely to result from a
random process. To further investigate this issue, we investigated the community structure of
the inferred connectivity, as well as the role of various internal parameters on the inference.
Finally, via simulations, we investigated the effect of the sparsity assumption on the ability of
the autoregressive model to recover ground-truth connectivity. The results of these analyses
are inconclusive, and suggest that in its current form, the Sparse Granger Causality
framework might not be the most optimal tool for large-scale connectivity estimation.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen