Part Two: Computational Neuroscience

In simple terms, computational neuroscience is the field of study where mathematical tools and theories are used to investigate brain function [1]. To understand how the nervous system processes information, it can also incorporate diverse approaches from electrical engineering, computer science and physics.

The brain is adaptive and dynamic, and neurons act in complex ways, making intuition and experimentation alone insufficient. To make sense of the large amounts of data from research on the brain, computational tools are necessary.

There are three types of approaches taken in computational neuroscience:

Currently, computational neuroscience is an extremely promising field, providing benefits to various areas of research. Brown University, for instance, is working on the BRAINSTORM program, aiming to bridge the gap between academia and industry in computational modeling and brain health [2]. BRAINSTORM investigates treatment for epilepsy, obsessive compulsive disorder (OCD), and biomarker discovery. Other exciting developments include modeling brain activity to assist with diagnosis for Parkinson’s disease, improving brain scan interpretation (fMRI and EEFG) to understand disorders, and decoding memory and learning processes to create new educational tools [3].

When it comes to computational neuroscience, there’s a lot that AI can assist with. Below are several real-world applications AI can provide in this field!

AI models are able to combine genetic, clinical, and imaging data in order to predict the progression of neurological diseases (as an example, see image below). Predictive models of Alzheimer’s and Parkinson’s are examples of this: deep learning algorithms simulate tau pathology spread in Alzheimer’s disease, while other models have shown motor systems’ relationship with dopamine depletion patterns [4]. This holds great promise for the development of precision medicine, contributing significantly to our understanding of complex brain processes.

Spiking neural networks are mathematical models of how neurons process information. They mimic the actual dynamics of biological neurons more closely, and are much more energy efficient compared to traditional neural networks.

SNNs have been used to model inhibitory–excitatory balance disruptions, which play a critical role in conditions like ASD and epilepsy. Through SNN simulations, altered synchronization in cortical circuits have been shown to contribute to cognitive impairments and sensory deficits [5]. Alongside this, SNNs leverage STDP (spike-timing dependent plasticity), which adjusts the connections between neurons based on the timing of their spikes [6]. STDP allows SNNs to learn from temporal patterns in data, similar to how we learn from experience over time. So, there’s great promise for AI models more adaptable and capable of lifelong learning, gradually improving their performance as they’re exposed to more data.

Besides these applications, it’s important to understand the limitations current models face. For instance, the process of training SNNs is extremely complex. Additionally, AI requires high-quality data, a barrier that neuroscience possesses due to lack of accessibility. Neuroimaging and clinical datasets could be influenced by outside factors, making AI models less reliable across diverse populations. Alongside this, AI requires strong computational resources, creating a challenge when adopting them in settings with limited resources [7].

Though computational systems have great potential, it is important to set clear goals in this field. The tools AI offers provides new opportunities, but also blurs the lines between neuroscience and similar disciplines. All in all, using AI models effectively requires combining large models and datasets while continuing to investigate interpretable theories of brain function.

Play this custom Connections game to check your understanding of computational neuroscience!