AI Model Unveils Insights into Neural Network Decline in ALS
New findings from a collaborative effort involving the University of St Andrews, the University of Copenhagen, and Drexel University have resulted in the creation of advanced AI computational models that forecast the deterioration of neural networks associated with Amyotrophic Lateral Sclerosis (ALS).
Published in the esteemed journal Neurobiology of Disease, this groundbreaking research opens the door for integrating computational modeling as a valuable tool alongside traditional animal studies and laboratory techniques.
Motor Neuron Disease (MND) encompasses a variety of disorders impacting motor neurons, which are vital nerve cells located in the brain and spinal cord. The most prevalent type is ALS, widely recognized by this name in numerous regions, but it is also referred to as Maladie de Charcot or Lou Gehrig's disease in different contexts. Globally, ALS affects about 2 individuals out of every 100,000 each year, translating to roughly 200 new diagnoses annually in Scotland alone.
Typically, ALS manifests first in the spinal region, leading to the initial impairment of motor neurons and specific neural circuits within the spinal cord. Early indications of the disease often include symptoms such as muscle weakness, stiffness, and cramps.
Historically, research on ALS has heavily relied on animal models, particularly genetically modified mice that exhibit symptoms akin to those of ALS patients. Researchers observe these animals to analyze the progression of the disease, but this approach often imposes constraints due to time and financial limitations, compelling scientists to focus only on certain stages of the disease's development.
Interestingly, computational models offer a unique advantage by simulating the processes occurring between these critical stages, thereby enhancing our understanding of disease progression. Moreover, these models enable researchers to replicate the same experiment while altering a single variable to determine its effect on the model's output, a luxury that traditional animal studies cannot afford due to the multitude of confounding factors present in living organisms.
Crucially, these computational frameworks empower researchers to predict how neural circuits may react to various treatments, providing valuable insights that can guide future preclinical trials involving mouse models.
The researchers utilized biologically plausible neural networks, which differ significantly from the conventional neural networks commonly used today for tasks like facial recognition on smartphones or engaging with AI chatbots like ChatGPT. These biologically inspired networks communicate through spike signals, emulating the way nerve cells interact within our nervous system. Their architecture is designed based on our current understanding of spinal cord cells and their interconnections, ensuring that the models reflect known biological realities.
The mathematical systems created by the team from the School of Psychology and Neuroscience compute the excitability levels of each neuron in the network. When a neuron receives a spike—an electrical impulse—its excitability shifts, and if this change crosses a certain threshold, the neuron will fire, transmitting information to the next neuron. To build the network, these neurons are categorized into populations, which are then interconnected based on biological evidence.
Beck Strohmer, a postdoctoral researcher at the University of Copenhagen and co-author of the study, explained, "In ALS, we know that neuron death occurs and that communication among neuron populations deteriorates. We replicate this in our models by systematically removing neurons from affected groups and diminishing the connections between them. This approach allows us to simulate the progression of the disease. Furthermore, we can explore treatment strategies aimed at preserving neurons or enhancing communication among them."
Dr. Ilary Alodi, another co-author and a Reader at the St Andrews School of Psychology and Neuroscience, emphasized the importance of validating hypotheses generated by models through animal testing, since capturing the full complexity of biological systems in a model is unattainable. "In this study, we predicted that our proposed treatment strategy would lead to the preservation of a specific group of neurons. Upon examining this neuron population in treated mice, we found our hypothesis was substantiated," she noted.
Such findings highlight that while predictions made by models should be interpreted with caution, they serve as excellent guides for experimental research.
This methodology not only enhances the precision of animal experiments but also equips researchers with clearer insights regarding where and when to monitor changes in these models.
Dr. Alodi further mentioned, "We are currently extending the application of these models to specific areas of the brain to investigate alterations in neuronal communication linked to dementia, marking an exciting new avenue for our research."
/Public Release. The content provided by the originating organization/author(s) may reflect a specific point in time and has been edited for clarity, style, and length. Mirage.News does not endorse any institutional positions or viewpoints; all interpretations and conclusions expressed here belong solely to the author(s). View the complete article here.
