New research from the University of St Andrews, the University of Copenhagen and Drexel University has developed AI computational models that predict the degeneration of neural networks in Amyotrophic Lateral Sclerosis (ALS).
Published in Neurobiology of Disease, the study paves the way to promote computational modelling as a complementary approach to the current animal and in vitro methods.
Motor neuron disease (MND) is the general name that's given to a group of illnesses which affect nerves called motor neurons in the brain and spinal cord. The most common subtype is ALS, which is also often the most widely used term in other countries. Other names include Maladie de Charcot and Lou Gehrig's disease. ALS affects approximately 2 out of 100,000 individuals per year globally. In Scotland this means approximately 200 people are diagnosed per year.
The majority of ALS cases show spinal onset, meaning that motor neurons and particular neural circuits in the spinal cord are affected first. This results in motor symptoms such as muscle weakness, muscle stiffness, and cramps being early signs of disease.
Traditionally, ALS is studied using animal models such as mice. The mice are genetically modified in order to have ALS-like symptoms. Their symptoms are then recorded to see how the disease progresses. Generally, when using animal models, researchers need to focus on specific timepoints during disease progression due to time and money constraints.
However, computational models can predict what is happening in between these timepoints to further understand disease progression. Additionally, they can repeat the same exact experiment with a single modification to understand the impact of a specific change to the output of the model whereas animals will always have many influencing factors.
Importantly, computational models also allow researchers to make predictions about how neural circuits may respond to treatment and they can inform future preclinical studies in mice.
Researchers in this study used biologically plausible neural networks. These networks are different compared to the traditional neural networks that are used every day for tasks like opening your smartphone with facial recognition or answering questions using ChatGPT. Biologically plausible neural networks communicate using spike signals, in a similar manner to the nerve cells found in our nervous system. The networks are structured based on the cells that scientists know exist in the spinal cord and how they are connected. In this way, the researchers develop their models based on what is known from the biology.
The models, developed by researchers from the School of Psychology and Neuroscience, are systems of mathematical equations that calculate the excitability of each neuron in the network. When a neuron receives a spike (an electrical impulse), this changes how excited the neuron is, and if it is excited enough, it will spike, thereby passing along the information to the next neuron. In order to construct the network, these neurons are grouped into populations and then the populations are connected based on biological data.
Co-author Beck Strohmer, postdoctoral researcher from the University of Copenhagen, said: "During ALS, it is known that neurons die and that the communication between populations break down. We model this by removing neurons from affected populations and by reducing the number of connections from affected populations. This allows us to model disease progression. In a similar way, we can model and test treatment strategies by saving neurons or strengthening communication."
Co-author Dr Ilary Alodi, Reader in St Andrews School of Psychology and Neuroscience, said: "Hypotheses generated by models need to be tested on animal models because it is impossible to model all the complexities of a biological system. In this study, we predicted that the applied treatment strategy in the model would save a specific population of neurons. We then looked at this neuron population in the treated mice and found that hypothesis held true."
Results like these show that whilst exercising caution with predictions from models, they are a good way to guide experimental research.
This also means that animal experimentation can be further refined because researchers have a better idea of where and when to look for changes in the animal models.
Dr Alodi added: "We are now starting to apply these models also to specific brain areas in order to understand how neuronal communication changes during dementia, which is an exciting new research direction for our lab."
