paper.md

title	tags	authors	affiliations	date	bibliography
Spectral Connectivity: a python package for computing multitaper spectral estimates and frequency-domain brain connectivity measures on the CPU and GPU	Python Neuroscience Multitaper analysis Spectral estimation Brain connectivity measures Fourier transform	name orcid affiliation Eric L. Denovellis 0000-0003-4606-087X 1, 2, 3 name affiliation orcid Maxym Myroshnychenko 4 0000-0001-7790-257X name orcid affiliation Mehrad Sarmashghi 0000-0002-7976-6636 5 name orcid affiliation Emily P. Stephen 0000-0003-1978-9622 6, 7	name index Howard Hughes Medical Institute - University of California, San Francisco, San Francisco, California, United States of America 1 name index Departments of Physiology and Psychiatry, University of California, San Francisco, San Francisco, California, United States of America 2 name index Kavli Institute for Fundamental Neuroscience, University of California, San Francisco, San Francisco, California, United States of America 3 name index National Institute of Neurological Disorders and Stroke, Bethesda, Maryland, United States of America 4 name index Division of Systems Engineering, Boston University, Boston, Massachusetts, United States of America 5 name index Department of Mathematics and Statistics, Boston University, Boston, Massachusetts, United States of America 6 name index Center for Systems Neuroscience, Boston University, Boston, Massachusetts, United States of America 7	10 September 2022	paper.bib

Summary

In neuroscience, characterizing the oscillatory dynamics of the brain is critical to understanding how brain areas interact and function. Neuronal activity tends to fluctuate rhythmically – both through intrinsic currents at the cellular level and through groups of neurons. Brain oscillations and their relationships can indicate the difference between normal and pathological brain states such as Alzheimer's and epilepsy. Spectral analysis techniques such as multitaper and wavelet analysis are widely used for decomposing signals into oscillatory components. Connectivity measures are used to determine the relationships between those oscillatory components, indicating possible communication between brain areas. Because these analyses are central to neuroscience and technological advances in recording are increasing the amount of simultaneously recorded signals, it is important to have a well-tested, standardized, and lightweight software package to compute these brain connectivity measures at scale.

Statement of Need

spectral_connectivity is a Python software package that computes multitaper spectral estimates and frequency-domain brain connectivity measures. The programming language Python is increasingly being used in the neurosciences [@MullerPythonNeuroscience2015; @SchlaflyPythonpracticingneuroscientist2020], but the two main packages for spectral analysis in Python, nitime [@Rokem2020] and mne-python [@GramfortEtAl2013a], have issues that make them more difficult to use in many situations. For example, nitime implements several estimators of the power spectrum, but lacks spectrograms and windowed spectral estimators. mne-python is a much larger package designed as a full-featured analysis library for EEG and MEG data, and works best when the data is represented using its ecosystem (i.e. Epochs and Raw objects). While some of the spectral connectivity functionality can work with array_like objects, users of other data modalities such as non-human electrophysiology data may find mne-python too cumbersome for their application. This is an important problem because the non-human neurosciences are undergoing a period of great technological development; more and more signals are being collected simultaneously, and the duration of these signals is becoming longer as chronic recordings become possible. This rapid increase in the size and duration of datasets demands a lightweight, fast, and efficient spectral estimation package. spectral_connectivity is designed to handle multiple time series flexibly¹ and can exploit GPUs for faster and more efficient computation. In addition, it can block compute important quantities such as the cross-spectral matrix in order to reduce memory burdens caused by large datasets. spectral_connectivity is also designed to be a lightweight package that has a simple user interface and can be easily be incorporated with other packages. Finally, spectral_connectivity also implements several connectivity measures that have not previously been implemented in Python such as the non-parametric version of the spectral granger causality and canonical coherence.

spectral_connectivity has already shown its utility to the neuroscience field. The package has already been used in a number of publications and pre-prints in neuroscience [@KuhnertDetectionDirectedConnectivities2019; @VargaNetworkPathConvergence2021; @LauroSubthalamicCorticalNetwork2021; @Delgado-SallentPhencyclidineinducedpsychosiscauses2021]. Interestingly, it has also contributed to a publication in physics [@CliffUnifyingPairwiseInteractions2022], showing its versatility and ease of use. We hope this package will continue to be useful to the neuroscience community, particularly for non-human electrophysiology data.

Acknowledgements

We thank Uri T. Eden for support and mentorship during the creation of this package.

Citations

Footnotes

1. For example, the expectation_type parameter of the Connectivity class gives the user the option to average over trials, tapers, or both, allowing the same module to yield single-trial or summarized estimates. ↩