How AI Learned to Listen to the Brain: From Guessing Words to Understanding How Neurons Work
Researchers at Peter the Great St. Petersburg Polytechnic University have developed an AI-based model capable of analyzing neural activity in the brain. According to the team, the breakthrough could help accelerate the development of new drugs for neurodegenerative diseases.
Artificial intelligence continues to push the boundaries of what can be studied and measured. Scientists at St. Petersburg Polytechnic University (SPbPU) have introduced a neural network that can analyze brain neuron activity with accuracy reaching 98%. The model relies on a principle familiar from modern language assistants: predicting what is missing. Instead of words, however, it reconstructs hidden fragments of electrical signaling between cells. The system is not yet a medical product, but it represents a fundamental advance that allows machines to begin genuinely "listening" to the language of neural activity.
The Architecture of a Neural Conversation
What exactly does the system do? The neural network was trained on a 270 GB dataset containing detailed recordings of neural activity in the visual cortex of laboratory mice. The algorithm learned to predict missing portions of the signal much like completing unfinished sentences. The most significant experimental result came later: after training, the model was successfully transferred to hippocampal data from the region responsible for memory formation. That suggests the AI captured not local noise patterns but broader principles underlying communication between neurons. For neuroscience, this kind of knowledge transfer marks a shift from studying isolated brain regions toward modeling the brain as a unified, dynamic system.
Why This Matters Strategically for Russia
The most immediate implication is a more precise understanding of the mechanisms behind brain disorders, faster drug discovery, and advances in diagnostics and personalized therapies. For patients with Alzheimer's disease, Parkinson's disease, and other neurological conditions, tools like this could eventually become part of a future development pipeline: understanding disease mechanisms → identifying candidate molecules → preclinical testing → clinical decision-making.
The project aligns closely with the global AI for Science movement. As Russia continues expanding its capabilities in digital medicine, the country gains its own advanced platform for analyzing biological signals. In practice, that could reduce dependence on foreign software while improving the reproducibility of domestic research.
The effort has been built systematically over several years. In 2023, researchers at St. Petersburg Polytechnic University released NeuroActivityToolkit (open-source software designed for statistical analysis of neural activity data). In 2025, they launched a dedicated laboratory focused on AI-driven analysis of biomedical images. Adapting language-model architectures to biological problems demonstrates that Russian research groups are not merely adopting existing algorithms but developing technological capabilities in a narrow yet strategically important field.
From Laboratory Mouse to Patient
Commercial deployment is not expected immediately, but a practical pathway is already emerging. The primary target is preclinical pharmaceutical research. Today, evaluation of potential treatments for neurodegenerative diseases often relies heavily on behavioral testing. The new AI model could allow researchers to observe, in real time, how neural connections reorganize in response to experimental compounds.
In 2025, SPbPU also opened a laboratory dedicated to AI-based analysis of biomedical images and data. Its mission includes studying neurodegenerative disorders such as Alzheimer's and Huntington's diseases. Parallel efforts are emerging across Russia, ranging from neural networks for early Parkinson's diagnosis using EEG data at Sechenov University to algorithms for monitoring psychophysiological states at institutes within the Russian Academy of Sciences. Over time, these initiatives could converge into a broader ecosystem for digital neurology.
AI as a Universal Translator Between Biology and Data
The new system developed by researchers in St. Petersburg is the result of five years of sustained work. From discussions surrounding the large-scale Mozg: zdorovye, intellekt, innovatsii (Brain: Health, Intelligence, Innovation) program to today's scientific publications, the direction has remained consistent: AI is becoming a universal translator between biology and digital systems. Researchers at SPbPU already plan to integrate animal behavioral telemetry into the model so that the algorithm can account not only for activity within the cortex but also for the behavioral context in which that activity occurs. If successfully scaled, the technology could evolve into an exportable scientific product - a software platform for preclinical research available to international pharmaceutical consortia.
Artificial intelligence is steadily moving beyond its role as a tool for automating routine tasks. It is becoming an interface for interacting with living systems. The Russian project illustrates how fundamental science, cross-cutting digital technologies, and clinical medicine can reinforce one another. For now, the work remains research code and laboratory data, but advances of this kind often represent the first steps toward future therapies and noninvasive diagnostic tools.
