Space-by-time semi-NMF: A decomposition approach for single-trial neural decoding

OBJECTIVE: The activity of populations of sensory neurons is assumed to encode stimulus information across both spatial and temporal dimensions, reflecting differences in tuning across neurons and variations in their responses over time. With the rapid expansion of simultaneously recorded neural populations in electrophysiological experiments, there is an increasing demand for analytical frameworks capable of compactly capturing this multidimensional sensory information within single-trial population activity, particularly under conditions of high baseline activity. APPROACH: To address this, we developed a novel analytical method that decomposes neural activity into spatial firing patterns (co-firing neuron ensembles), temporal firing patterns (temporal dynamics of these groups of neurons), and trial-dependent activation coefficients (strength of recruitment of such firing patterns on each trial). This method extends the space-by-time nonnegative matrix factorization (SbT-NMF), a tensor decomposition that imposes nonnegativity constraints on all factorized matrices and is suitable for inherently nonnegative data such as neural activity by relaxing the nonnegativity constraint on one matrix. This modification enables the application of an important baseline correction, which involves subtracting prestimulus activity from the data. As a result, the input may contain negative values, reflecting deviations from baseline and thereby reducing the influence of ongoing activity before the stimulus. MAIN RESULTS: We applied this refined method to primary visual cortex data from the Allen Brain Institute and found that SbT-sNMF (66% ± 0.7%) achieved a 32% relative increase in stimulus decoding performance over the original SbT-NMF (50% ± 1.4%). SIGNIFICANCE: These findings highlight the importance of accounting for baseline correction to achieve compact and linearly decodable low-dimensional representation of neural population dynamics in our proposed frameworks.

Authors

Institutions

Publication Details

Journal
Journal of Neural Engineering
Published
2026-11-16
DOI
https://doi.org/10.1088/1741-2552/aec2a9
Primary Topic
Tensor decomposition and applications
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Space-by-time semi-NMF: A decomposition approach for single-trial neural decoding

Fleur Zeldenrust, Arezoo Alizadeh
Journal of Neural Engineering
Tensor decomposition and applications
article

Space-by-time semi-NMF: A decomposition approach for single-trial neural decoding

Fleur Zeldenrust, Arezoo Alizadeh
article en

Abstract

OBJECTIVE: The activity of populations of sensory neurons is assumed to encode stimulus information across both spatial and temporal dimensions, reflecting differences in tuning across neurons and variations in their responses over time. With the rapid expansion of simultaneously recorded neural populations in electrophysiological experiments, there is an increasing demand for analytical frameworks capable of compactly capturing this multidimensional sensory information within single-trial population activity, particularly under conditions of high baseline activity. APPROACH: To address this, we developed a novel analytical method that decomposes neural activity into spatial firing patterns (co-firing neuron ensembles), temporal firing patterns (temporal dynamics of these groups of neurons), and trial-dependent activation coefficients (strength of recruitment of such firing patterns on each trial). This method extends the space-by-time nonnegative matrix factorization (SbT-NMF), a tensor decomposition that imposes nonnegativity constraints on all factorized matrices and is suitable for inherently nonnegative data such as neural activity by relaxing the nonnegativity constraint on one matrix. This modification enables the application of an important baseline correction, which involves subtracting prestimulus activity from the data. As a result, the input may contain negative values, reflecting deviations from baseline and thereby reducing the influence of ongoing activity before the stimulus. MAIN RESULTS: We applied this refined method to primary visual cortex data from the Allen Brain Institute and found that SbT-sNMF (66% ± 0.7%) achieved a 32% relative increase in stimulus decoding performance over the original SbT-NMF (50% ± 1.4%). SIGNIFICANCE: These findings highlight the importance of accounting for baseline correction to achieve compact and linearly decodable low-dimensional representation of neural population dynamics in our proposed frameworks.

Journal of Neural Engineering
Donders Institute for Brain, Cognition and Behaviour (NL)
Nederlandse Organisatie voor Wetenschappelijk Onderzoek
Tensor decomposition and applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon