PhysMAP - interpretablein vivoneuronal cell type identification using multi-modal analysis of electrophysiological data

Abstract

Cells of different types perform diverse computations and coordinate their activity during sensation, perception, and action. While electrophysiological approaches can measure the activity of many neurons simultaneously, assigning cell type labels to these neurons is an open problem. Here, we develop PhysMAP, a framework that weighs multiple electrophysiological modalities simultaneously in an unsupervised manner and obtain an interpretable representation that separates neurons by cell type. PhysMAP is superior to any single electrophysiological modality in identifying neuronal cell types such as excitatory pyramidal, PV+interneurons, and SOM+interneurons with high confidence in both juxtacellular and extracellular recordings and from multiple areas of the mouse brain. PhysMAP built on ground truth data can be used for classifying cell types in new and existing electrophysiological datasets, and thus facilitate simultaneous assessment of the coordinated dynamics of multiple neuronal cell types during behavior.

A conserved code for anatomy: Neurons throughout the brain embed robust signatures of their anatomical location into spike trains

“…Another major axis along which cortical neurons exhibit anatomical structure is layer. This aligns with recent evidence that indicates that cortical layers can be distinguished computationally, particularly in terms of cell types and tuning preferences [15,45,46]. Thus, we hypothesized that cortical layer might be more reliably embedded in single unit spike times than visuocortical structure.…”

Ultra-high density electrodes improve detection, yield, and cell type identification in neuronal recordings

“…In fact, our category of 'untagged' units includes neurons from the PV, VIP, and SST cell classes, which likely leads to an underestimation of decoding accuracy (i.e., some untagged units labeled by the classifier as VIP might in fact be VIP+, though this is counted as an error). In the future, unit classification using a variety of additional metrics, such as non-linear waveform features (Lee et al, 2021), autocorrelograms (Barthó et al, 2004;Yamin et al, 2013;Senzai and Buzsáki, 2017), and oscillatory phase locking (Reifenstein et al, 2016), as well as alternative machine learning approaches (Crawford and Pineau, 2019;Tymochko et al, 2020;Kingma and Welling, 2022;Beau et al, 2024;Lee et al, 2024), may yield even greater performance.…”

“…Finally, the brain contains a large diversity of cell types, and extracellular electrophysiology has been limited in its ability to discriminate between these types (Gouwens et al, 2020;Yao et al, 2021). Classically, extracellularly recorded units in some brain regions such as cortex and striatum have been separated into coarse cell types on the basis of features such as waveform shape and firing pattern (McCormick et al, 1985;Barthó et al, 2004;Mitchell et al, 2007;Niell and Stryker, 2008;Yamin et al, 2013;Roux et al, 2014;Senzai and Buzsáki, 2017;Yu et al, 2019;Lee et al, 2021Lee et al, , 2024Beau et al, 2024). NP 1.0 probes can provide additional coarse morphological information about the electrical field surrounding a neuron useful for neuron classification (Buccino et al, 2018;Jia et al, 2019), but this information may not be detailed enough to resolve the large diversity of cell types in the brain.…”

Circuit dynamics of superficial and deep CA1 pyramidal cells and inhibitory cells in freely-moving macaques

“…Binary classification is limited in fully capturing the structural and transcriptomic diversity in monkeys and mice 9,89 , and recent findings challenge the adequacy of this approach 90–98 . In addition to refinements to waveform classification 59,90,99,100 , cells’ firing patterns may prove vital for more accurate and reliable classification 101–103 . The firing patterns and timing of neuron activity summarized by the inter-spike interval (ISI) distributions have previously been applied to differentiate inhibitory from excitatory cells across brain regions and species 52,60,66,104–109 .…”

“…One caveat to this approach is that unsupervised methods, which excel at capturing diversity, do not automatically ensure consistency across studies. Thus, future research should investigate how physiological diversity reported using nonlinear unsupervised methods co-vary with morphological and transcriptomic features on a cell-by-cell basis to understand the relationships among these three modalities as they pertain to cell-type definition or cell state-dependent variations (see 103,110 ). This will enable a recalibration of classification criteria in accordance with the emerging differences in morphology and physiology across species and brain structure.…”