In Real Time: AI Identifies Neurons By Their “Electrical Signal” With 95% Accuracy

Scientists have developed an artificial intelligence system that can identify with over 95% accuracy which types of neurons are active during behaviors, simply by analyzing electrical signals from the brain. Using techniques such as optogenetics and deep learning, the system recognizes specific patterns of electrical activity and brain cell anatomy. This advance could help us better understand how different neurons work together and accelerate research into the brain and neurological diseases.

The brain is made up of a huge variety of neurons, each with specific functions, shapes, and properties. Understanding which types of neurons are active during certain tasks or behaviors is essential to deciphering how the brain works.

However, identifying these cell types accurately, especially in recordings made in living, awake animals, is a major challenge.

Scientists typically use electrophysiological probes that record the electrical activity of many neurons at the same time, but these probes cannot directly reveal the type of cell involved, but only show the “spikes” of electrical activity.

To solve this problem, researchers developed a new artificial intelligence (AI)-based method capable of automatically identifying the type of neuron based on data from these electrical recordings.

Electrophysiological probe recording electrical activity of a single neuron

They used the cerebellum, a part of the brain important for motor control and learning, as a model. The study focused on four main types of cerebellar cells: Purkinje cells, molecular layer interneurons, Golgi cells and mossy fibers.

How was this done? First, the scientists created a "reference library" of real data, accurately identifying the type of each neuron. This was achieved using two advanced techniques:

1. Optogenetics: a technique that uses light to activate specific genetically modified neurons. This allows them to identify precisely which cell is responding.

   
   

2. Synaptic blockade: which helps isolate the activity of individual neurons, eliminating interference from connections with other cells.

With this information, detailed electrical data were collected, such as the shape of the wave generated by the neuron (waveform), the frequency and pattern of firing (firing statistics), and the anatomical layer where the neuron is located.

This data was then used to train a deep learning classifier, a type of neural network that learns complex patterns in large data sets. This classifier was trained in a semi-supervised manner, that is, with some of the data labeled and some unlabeled, which better reflects real-world scenarios in the experiments.

An example of a neural network that learns complex patterns from large data sets in a semi-supervised manner, Image: Ramin Hasani/MIT CSAIL using the stable diffusion AI image generator.

The system learned to recognize unique combinations of features that identify each type of neuron with over 95% accuracy.

Most impressively, the classifier performed with high accuracy even in different labs, using different types of probes, in different regions of the cerebellum, and even across different species, such as mice and monkeys.

In addition to identifying cell types, the AI ​​allowed scientists to observe how different types of neurons behave differently during tasks, opening new windows into understanding the specific functions of each cell in brain processing.

This breakthrough represents a powerful tool for modern neuroscience. With the help of AI, it will be possible to analyze the activity of entire brain circuits in much greater detail and better understand how the brain works in real time, something essential for the advancement of neurological treatments and the development of brain-machine interfaces.

Diagram shows how researchers used optogenetics, synaptic gating, and large-scale brain recordings to identify specific types of neurons. This data was used to train an artificial intelligence that recognizes electrical and anatomical features of cells. The system can predict neuron type with high accuracy across species and brain regions, revealing how each cell type contributes to brain dynamics during behavior.

READ MORE:

A deep learning strategy to identify cell types across species from high-density extracellular recordings

Maxime Beau, David J. Herzfeld, Francisco Naveros, Marie E. Hemelt, Federico D’Agostino, Marlies Oostland, Alvaro Sánchez-López, Young Yoon Chung, Michael Maibach, Stephen Kyranakis, Hannah N. Stabb, M. Gabriela Martínez Lopera, Agoston Lajko, Marie Zedler, Shogo Ohmae, Nathan J. Hall, Beverley A. Clark, Dana Cohen, Stephen G. Lisberger, Dimitar Kostadinov, Court Hull, Michael Häusser, Javier F. Medina

Cell. Volume 188, Issue 8P2218-2234.E22April 17, 2025

DOI: 10.1016/j.cell.2025.01.041

Abstract:

AI Learns to Decode Neuron Types From Brain Signals With 95% AccuracyHigh-density probes allow electrophysiological recordings from many neurons simultaneously across entire brain circuits but fail to reveal cell type. Here, we develop a strategy to identify cell types from extracellular recordings in awake animals and reveal the computational roles of neurons with distinct functional, molecular, and anatomical properties. We combine optogenetics and pharmacology using the cerebellum as a testbed to generate a curated ground-truth library of electrophysiological properties for Purkinje cells, molecular layer interneurons, Golgi cells, and mossy fibers. We train a semi-supervised deep learning classifier that predicts cell types with greater than 95% accuracy based on the waveform, discharge statistics, and layer of the recorded neuron. The classifier’s predictions agree with expert classification on recordings using different probes, in different laboratories, from functionally distinct cerebellar regions, and across species. Our classifier extends the power of modern dynamical systems analyses by revealing the unique contributions of simultaneously recorded cell types during behavior.