Unsupervised discovery of temporal sequences in high-dimensional datasets, with applications to neuroscience

Abstract

Identifying low-dimensional features that describe large-scale neural recordings is a major challenge in neuroscience. Repeated temporal patterns (sequences) are thought to be a salient feature of neural dynamics, but are not succinctly captured by traditional dimensionality reduction techniques. Here, we describe a software toolbox—called seqNMF—with new methods for extracting informative, non-redundant, sequences from high-dimensional neural data, testing the significance of these extracted patterns, and assessing the prevalence of sequential structure in data. We test these methods on simulated data under multiple noise conditions, and on several real neural and behavioral data sets. In hippocampal data, seqNMF identifies neural sequences that match those calculated manually by reference to behavioral events. In songbird data, seqNMF discovers neural sequences in untutored birds that lack stereotyped songs. Thus, by identifying temporal structure directly from neural data, seqNMF enables dissection of complex neural circuits without relying on temporal references from stimuli or behavioral outputs.

Unsupervised clustering of temporal patterns in high-dimensional neuronal ensembles using a novel dissimilarity measure

“…It is important to emphasize that our method is an explicit clustering method, that can find unique patterns of network activity that are well separated from one another. Several methods using decomposition techniques like PCA or matrix factorization have been utilized with the goal of extracting patterns or sequences from neuronal ensemble data (Peyrache et al, 2009;Mackevicius et al, 2018;Lopes-dos-Santos, Ribeiro, and Tort, 2013;Stopfer, Jayaraman, and Laurent, 2003). We highlight several differences between SPOTDisClust and decomposition techniques: 1) In principle, decomposition techniques like PCA achieve a different goal, namely to identify components that explain a large fraction of variance in the data.…”

Routing of non-mnemonic hippocampal ripples to subcortical structures

“…As the proportion of hippocampal SPW-Rs that can be decoded as 'replay' events is small in the hippocampus (8-20%; Davidson; other REFS), we reasoned that perhaps some independent sequential structure in other hippocampal SPW-Rs may drive LS neuron firing. To examine this possibility, we took an unsupervised variance decomposition approach (Mackevicius et al, 2019). This method allowed us to extract sequential activity patterns from hippocampal SPW-Rs without relying on a pre-existing template.…”

“…A recently developed unsupervised method of spatiotemporal quantification of neural populations was also used (Mackevicius et al, 2019). This method allows for the examination of consistently occurring neural sequences without relying on a behavioral 'template' to relate too.…”

Strongly correlated spatiotemporal encoding and simple decoding in the prefrontal cortex

“…Our analysis of population dynamics through modeling of spiking patterns using fine temporal windows, brings together the power of decomposition of spatio-temporal structure of population activity and detailed modeling of spiking patterns of large neural populations. This combination extends methods like tensor decomposition [63,64], as we provide an interpretable generative model and do not depend on metric based approaches, and machine learning based approaches for modeling spiking patterns as autoencoders (LFADS). More broadly, the current framework is easily applicable to other neural systems, and can be used to study the nature of sequences of states [18,19,65].…”