Summary: Recording the electrical “noise” of the brain has long been possible, but understanding which specific cells are doing the talking has remained a major hurdle. Researchers have developed PhysMAP, a machine learning tool that can distinguish between different types of neurons based solely on their unique electrical signatures.
The study demonstrates that PhysMAP can identify cells implicated in psychiatric disorders, such as those linked to schizophrenia and major depressive disorder, directly from in vivo recordings. This breakthrough allows scientists to study how specific neural circuits fail in real-time, providing a roadmap for targeted, next-generation psychiatric therapies.
Key Facts
- Solving Circuitopathies: PhysMAP is designed to identify “circuitopathies”—disorders like schizophrenia and depression that stem from dysfunctional interactions between specific cell types rather than overall brain activity.
- A “Voice” Separator: The machine learning algorithm identifies specific neuron types (like parvalbumin or somatostatin cells) by combining multiple electrical signatures, allowing researchers to isolate individual “voices” within a crowded recording.
- No Genetic Engineering Needed: Unlike previous methods that required complex “optotagging” or genetic manipulation, PhysMAP can identify cell types in living brains using only electrical recordings.
- Leveraging Open Data: The tool was trained and validated using seven public datasets, proving that open-source scientific data can be repurposed to create powerful new diagnostic technologies without new experiments.
Source: Boston University
When probes are inserted into the brain for research or clinical purposes, the electrical activity of neurons is recorded. These signals can be used to understand how the brain performs certain computations or even to identify pathological states.
However, brains are composed of cell types that perform different roles in computation and are differentially affected by certain psychiatric disorders or drugs. Without a deep understanding of how cell types orchestrate the overall activity patterns, we cannot develop the next generation of therapies.
Researchers from Boston University’s Chobanian & Avedisian School of Medicine, College of Arts & Sciences, College of Engineering and Faculty of Computing & Data Sciences have developed a tool called PhysMAP to separate the “voices” of individual cell types within a crowd of electrical noise by combining several complementary features of each type’s electrical signature.
This machine learning algorithm could open up the study of how cell types shape both the healthy computations and the pathological states that electrical recordings have long been able to detect but never fully understood.
“While a variety of disorders can be understood via overall activity alone, a growing number of psychiatric disorders are being recognized as arising from the perturbed interactions between specific cell types rather than changes in overall activity. These so-called “circuitopathies” include human diseases such as schizophrenia, major depressive disorder, and some forms of epilepsy.
PhysMAP would allow for the study of interacting cell types in both intact and altered neural circuits, many implicated in the above disorders, in expanded in vivo research settings and perhaps even clinical ones,” explains corresponding author Chandramouli (Chand) Chandrasekaran, PhD, assistant professor of anatomy & neurobiology and psychological and brain sciences at BU.
The researchers used seven open datasets that contained both the electrical activity of single neurons and their cell type identities. In these experiments, scientists combined molecular engineering with optical tools to tie electrical activity to specific cell types with a technique called “optotagging;” and released these datasets when their papers were published.
The BU team used these datasets to train PhysMAP to learn the unique electrical signatures of different cell types and verify that this mapping was accurate and better than or comparable to other tools.
Crucially, once learned, this mapping could be applied to new datasets where optotagging was not available, enabling the simultaneous study of multiple cell types. This work also illustrates the power of open data sharing. By making their datasets publicly available, scientists enabled the development and validation of entirely new tools without requiring additional experiments.
According to the researchers, the ability to study cell types in vivo—without the requirement of genetic manipulation—would allow the study of how psychiatric disorders arise from circuit dysfunction. “If these cell types can be identified in research settings in the healthy brain, information about their dysfunction can be used to inform the development of future therapeutic strategies,” adds Chandrasekaran.
A previous version of this tool (WaveMAP), was deployed to identify cell types in the very first human recordings with Neuropixels (a type of high-density electrode now considered the standard in neuroscience).
PhysMAP is more powerful and can be used to identify several of the specific cell types implicated in psychiatric disorders: parvalbumin-positive cells in schizophrenia or Dravet syndrome and somatostatin-positive cells in major depressive disorder.
These findings appear online in the journal Nature Communications.
Funding: CC was supported by an NIH NINDS R00NS092972, R01NS121409, R21NS135361 and R01NS122969 award; the Moorman-Simon Interdisciplinary Career Development Professorship from Boston University; the Whitehall Foundation (2019-12-77); and the Young Investigator Award from the Brain and Behavior Research Foundation (27923).
The auditory cortex dataset (collected by AL and SJ) was supported by an NIH NIDCD R01DC01553. SJ was also supported by an NIH NINDS RF1NS131993. EKL was supported by an NIH NINDS F31NS131018. The Neuropixels Ultras dataset was supported by NIH NINDS/NIMH U01NS113252 awarded to NS.
Key Questions Answered:
A: Brain activity is like a stadium crowd cheering; you can hear the volume, but you don’t know who is starting the chant. Disorders like schizophrenia or epilepsy aren’t just about “too much” or “too little” activity, they happen when specific types of cells (like inhibitory parvalbumin cells) stop talking to their neighbors correctly. PhysMAP lets us listen to those specific cells.
A: The team used “optotagging” data, where light is used to verify a cell’s identity—as a training manual. PhysMAP learned the subtle timing, shape, and frequency of the electrical pulses unique to each cell type. Now, it can recognize those “signatures” in new recordings where no light or tags are present.
A: That is the ultimate goal. Because PhysMAP doesn’t require genetic manipulation, it could eventually be used alongside standard clinical electrodes (like Neuropixels) to help doctors diagnose the specific cellular cause of a patient’s psychiatric symptoms and choose a more effective drug.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this AI mental health research news
Author: Gina DiGravio
Source: Boston University
Contact: Gina DiGravio – Boston University
Image: The image is credited to Neuroscience News
Original Research: Open access.
“A multimodal approach for visualizing and identifying electrophysiological cell types in vivo” by Eric Kenji Lee, Asım E. Gül, Greggory Heller, Anna Lakunina, Han Yu, Andrew Shelton, Shawn Olsen, Nicholas A. Steinmetz, Cole Hurwitz, Santiago Jaramillo, Pawel F. Przytycki & Chandramouli Chandrasekaran. Nature Communications
DOI:10.1038/s41467-026-71331-0
Abstract
A multimodal approach for visualizing and identifying electrophysiological cell types in vivo
Neurons of different types perform diverse computations and coordinate their activity during sensation, perception, and action. While electrophysiological recordings can measure the activity of many neurons simultaneously, identifying cell types during these experiments remains difficult.
Here we present PhysMAP, a framework adapted from multiomics data analysis that weights multiple electrophysiological modalities simultaneously to obtain interpretable multimodal representations.
We apply PhysMAP to seven datasets and demonstrate that these multimodal representations are better aligned with known transcriptomically-defined cell types than any single modality alone. We then show that this alignment allows PhysMAP to better identify putative cell types in the absence of ground truth.
We also demonstrate how annotated datasets can transfer labels to unannotated recordings and confirm that inferred cell types exhibit properties consistent with ground truth. Crucially, we show that PhysMAP can also be used to iteratively detect batch effects which confound classification.
Together, these results establish PhysMAP as a tool for studying multiple cell types simultaneously and gaining insight into neural circuit dynamics.
