Neuroscientists Use AI to Decode How Brain Forms New Memories

Neuroscientists have developed an AI tool that offers unprecedented insights into the cerebellum, a key brain region responsible for precise movements. Despite long-standing methods to record and monitor the electrical signals from neurons in the cerebellum, the computational processes between these signals remained largely unknown. However, a collaborative effort involving 23 scientists from various institutions, including Baylor College of Medicine, Duke University, University College London, University of Granada, University of Amsterdam, Bar-Ilan University, and King's College London, has created a groundbreaking solution. The research, published in the journal *Cell*, introduces a semi-supervised deep learning classifier. This tool can identify the types of neurons from electrical signals recorded during behavior, effectively deciphering which "language" each neuron is "speaking" in the cerebellar network. Dr. Jose Medina, a Brown Foundation Professor and director of the Neuroscience and AI Center at Baylor College of Medicine, likened the process to distinguishing multiple languages in a crowded room. "Our new AI tool can recognize the unique electrical signature of each neuron, allowing us to classify them," Medina explained. This breakthrough is particularly significant because it marks the first step in understanding the content of neural conversations, a crucial aspect in elucidating how the brain generates behavior. Dr. Stephen Lisberger, one of the seven senior authors from Duke University, reflected on the challenge: "For years, we couldn't explain how input signals were transformed into output signals. Advanced recording techniques could capture electrical activity, but not the neuron types responsible. This discovery is a turning point, potentially opening new doors for treating neurological disorders." To build the classifier, the team first measured the distinct electrical properties of different cerebellar neuron types using optogenetics. This technique involves introducing light-sensitive proteins to specific neurons to "label" their electrical activity. The data from these measurements were used to train the deep learning classifier, enabling it to categorize recorded neural signals by neuron type. Dr. David Herzfeld, a senior research associate at Duke University and one of the seven lead authors, emphasized the significance: "This tool is a major advancement in our understanding of cerebellar information processing. We hope it will inspire similar tools for other brain regions, ultimately leading to new treatments for neurological conditions." Industry insiders view this development as a critical milestone in neuroscience, solving the neuron type identification problem and laying the foundation for understanding neural communication. The research not only aids in comprehending cerebellar functions but also promises to advance treatments for various neurological disorders. Baylor College of Medicine is renowned for its research in neuroscience and artificial intelligence, and this collaboration with leading international institutions further solidifies its position at the forefront of these fields. Other participating institutions, such as Duke University, also bring extensive research experience and achievements. --- Understanding how the brain forms and retains new memories is a fundamental question in neuroscience. A new study published in the journal *Science* has shed light on some of the "rules" the brain follows during the learning process. The human brain consists of billions of neurons that communicate through electrical signals, similar to how computers use binary code. These signals are transmitted via synapses, the connections between neurons, and received by dendrites, the branching structures of neurons that integrate thousands of signals. Scientists have long hypothesized that brain learning involves altering the connections between neurons. New information and experiences change the communication patterns and collective activity of neurons, strengthening some synapses and weakening others. This synaptic plasticity process is crucial for storing new information and experiences. However, the mechanisms by which the brain decides which synapses to strengthen or weaken—the credit assignment problem—have been elusive. To address this, researchers monitored the activity of individual synapses in the brains of mice, observing which patterns of activity determine the strengthening or weakening of connections. They implanted biosensors in the mice's neurons, which lit up when synaptic and neural activities occurred. The mice were trained to press a lever to a specific position after hearing a sound to receive water, a task that required new learning. The findings were surprising: not all synapses on the same neuron followed the same rules. Some adhered to the Hebbian rule, where consistent electrical activity between neurons strengthens their connection, while others showed different behaviors, remaining unaffected by neuronal activity. This suggests that neurons can regulate different types of inputs more precisely by using multiple learning rules simultaneously, enhancing their ability to encode new information. This discovery deepens our understanding of how synapses change during learning. Given that most brain disorders, including degenerative and psychiatric conditions, involve some form of synaptic dysfunction, the findings have potential implications for human health. For instance, depression might result from excessive weakening of synapses in certain brain areas, making it difficult to experience pleasure. By elucidating the normal mechanisms of synaptic plasticity, scientists can better diagnose and treat such conditions. The research also has implications for artificial intelligence. Current artificial neural networks, which underpin AI models, often use uniform learning rules that do not align closely with biological principles. This study could inspire the development of more biologically plausible AI models with enhanced performance and efficiency. Future research will delve into the specific causes and mechanisms of these learning rules and explore the additional properties conferred by multitasking capabilities in neurons. While it is clear that neurons can use multiple methods to improve information encoding, their full range of functions remains an open question. Experts see this finding as a new perspective on brain learning, with the potential to drive the development of computational models that incorporate multitasking capabilities. The researchers, hailing from top universities, have provided crucial insights into the mysteries of memory formation and neural learning. The development of this AI tool and the new understanding of synaptic plasticity represent significant advancements in neuroscience, offering promising avenues for both medical and technological applications. These findings not only enhance our knowledge of brain function but also pave the way for more effective treatments of neurological and psychiatric disorders and the creation of more sophisticated AI models.