Summary: Recording the brain’s electrical “noise” has long been possible, but determining which specific cells generate those signals has remained a major challenge. Researchers have developed PhysMAP, a machine learning framework that can distinguish neuron types from their unique electrical signatures. This approach enables identification of cell types implicated in psychiatric disorders directly from in vivo recordings, opening a path to study circuit-level dysfunctions in real time and guiding targeted future therapies.
PhysMAP analyzes multiple complementary features of neuronal electrical activity to separate the individual “voices” inside complex extracellular recordings. The tool detects signatures of cell classes such as parvalbumin-positive and somatostatin-positive neurons—cell types implicated in conditions like schizophrenia, Dravet syndrome, and major depressive disorder—without requiring genetic tagging or optical labeling.
Key Facts
- Addressing circuitopathies: PhysMAP is aimed at identifying disorders that arise from dysfunctional interactions among specific neuron types—so-called circuitopathies—rather than global changes in overall brain activity.
- Separating neuronal “voices”: The algorithm combines timing, waveform, and frequency features to recognize the distinct electrical fingerprints of different cell types recorded together on high-density probes.
- No genetic manipulation required: Unlike optotagging-based approaches, PhysMAP can classify cell types from standard extracellular recordings, enabling broader application in experimental and potentially clinical settings.
- Built from open data: The method was trained and validated on seven publicly released datasets that paired electrophysiological recordings with cell-type identities, demonstrating how shared data accelerates new tool development without additional experiments.
Source: Boston University
When electrodes are placed in the brain for research or clinical monitoring, they capture the electrical activity of many neurons at once. These recordings reveal how neural populations compute and can signal pathological states. However, the brain contains many distinct cell types that play unique computational roles and are differently affected by psychiatric disorders and treatments. Without knowing which cell types are active or failing, it is difficult to design precise interventions.
A multidisciplinary team at Boston University—including researchers from the Chobanian & Avedisian School of Medicine, College of Arts & Sciences, College of Engineering, and Faculty of Computing & Data Sciences—developed PhysMAP to map electrophysiological recordings onto interpretable, multimodal representations of cell type. By weighting and integrating multiple electrophysiological features, PhysMAP isolates the signals of individual neuron classes from crowded in vivo recordings.
PhysMAP was trained on seven open datasets that included both extracellular activity and independently validated cell-type labels. Those ground-truth labels were obtained in the original experiments using molecular and optical methods—commonly through optotagging—then shared publicly with their publications. The Boston University team used these labeled datasets to teach PhysMAP the characteristic timing, shape, and frequency patterns associated with specific neuron classes and to verify that the resulting classifications matched or exceeded the performance of other approaches.
Importantly, once PhysMAP learns these multimodal electrophysiological signatures, it can transfer that knowledge to new, unlabeled recordings where optotagging is not available. This capability enables simultaneous study of multiple cell types in intact neural circuits and highlights the value of open data sharing: previously published datasets served as the training foundation for an entirely new diagnostic technology, without requiring additional animal or human experiments.
PhysMAP builds on earlier tools (for example, WaveMAP was applied to early human Neuropixels recordings) but extends them to provide more robust identification of cell types implicated in psychiatric disorders—specifically, parvalbumin-positive interneurons and somatostatin-positive interneurons—using only electrophysiological signals.
“A growing body of evidence suggests that many psychiatric conditions arise from disrupted interactions between particular cell types rather than overall increases or decreases in activity,” says corresponding author Chandramouli (Chand) Chandrasekaran, PhD, assistant professor of anatomy & neurobiology and psychological and brain sciences at BU. “PhysMAP makes it possible to study those interacting cell types in vivo without genetic manipulation, which could inform development of targeted therapeutic strategies.”
The study’s findings are published in the journal Nature Communications.
Funding: CC was supported by NIH NINDS grants R00NS092972, R01NS121409, R21NS135361 and R01NS122969; the Moorman-Simon Interdisciplinary Career Development Professorship from Boston University; the Whitehall Foundation (2019-12-77); and a Young Investigator Award from the Brain & Behavior Research Foundation (27923). The auditory cortex dataset (collected by AL and SJ) was supported by NIH NIDCD R01DC01553. SJ received additional support from NIH NINDS RF1NS131993. EKL was supported by NIH NINDS F31NS131018. The Neuropixels Ultras dataset received funding from NIH NINDS/NIMH U01NS113252 awarded to NS.
Key Questions Answered:
A: Electrophysiological recordings tell us the overall level and timing of activity—like hearing the roar of a stadium crowd—but not which individuals are driving specific patterns. Many disorders, including some forms of epilepsy and psychiatric illnesses such as schizophrenia, arise when particular cell types fail to interact correctly with their partners. Identifying those cell types helps pin down circuit-level causes of dysfunction and suggests more precise interventions.
A: The researchers trained PhysMAP on datasets where optotagging or other molecular labeling confirmed cell identity. Using those labeled examples as a guide, the algorithm learned which combinations of waveform shape, firing timing, and frequency content correspond to particular cell classes. Once trained, PhysMAP can detect those multimodal signatures in new recordings without labels.
A: Translating PhysMAP to clinical use is an important long-term goal. Because the method relies on extracellular electrophysiology and not genetic manipulation, it could in principle be applied to high-density clinical probes to help diagnose the cellular origins of a patient’s symptoms and inform treatment selection. Further validation in human recordings will be required.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full by the editorial team.
- Additional context and clarifications were added by 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
Different neuron types perform distinct computations and coordinate their activity during sensation, perception, and action. Although electrophysiological recordings can capture the activity of many neurons simultaneously, identifying cell types within those recordings is challenging. PhysMAP is a framework inspired by multiomics analysis that integrates and weights multiple electrophysiological modalities to produce interpretable multimodal representations.
Applied to seven datasets, PhysMAP yields multimodal representations that align more closely with transcriptomically defined cell classes than any single electrophysiological modality alone. This alignment improves identification of putative cell types when ground truth labels are absent, enables label transfer from annotated to unannotated recordings, and produces inferred cell types with properties consistent with verified identities. The framework also supports iterative detection of batch effects that can confound classification across datasets.
Together, these results establish PhysMAP as a practical tool for simultaneously studying multiple cell types in in vivo electrophysiological recordings and for gaining insight into neural circuit dynamics and their role in health and disease.