Modelling Long-Term Human Brain Activity: Are models still valid given errors in measurement?

Abstract

Whole Brain Emulation (WBE) represents one of the most ambitious objectives in contemporary computational neuroscience, aiming to replicate human brain activity within a computational model. This study investigates the role of measurement errors during the data acquisition phase and their subsequent impact on neural simulations, focusing on the Kuramoto and Izhikevich models. Both models were employed to simulate the dynamics of different brain regions, particularly focusing on the introduction of noise that mimics errors originating from brain imaging techniques. Our analysis begins with the observation of how noise affects neural dynamics by segmenting the simulations into three phases: 1. The control phase before noise introduction, representing a brain’s natural state. 2. The branching point, where noise is introduced as a representation of data acquisition errors in WBE. 3. The simulation of both control data (undisturbed brain function) and noisy data (the behavior of a brain replica impacted by measurement errors). A key finding of this work is the clear correlation between noise levels and the total error in both models, confirming that higher noise results in greater error. This underscores the critical importance of using precise measurement techniques during the data acquisition phase and suggests the need for developing error-correction mechanisms to mitigate the impact of noise. We also investigated the impact of connectivity strength in specific brain regions, revealing distinct differences between the models. In the Kuramoto model, regions with higher connectivity contributed more to the final error, while in the Izhikevich model, these same regions tended to reduce error as their connectivity increased. These findings are significant because they highlight the need for further investigation into measurement error in WBE, in addition to ongoing work on computational and hardware aspects of brain emulation. As demonstrated, noise introduced by data acquisition has a profound impact on neural simulation accuracy, and addressing this challenge is essential for achieving reliable WBE. Additionally, while the models employed in this study lack certain biological realism, including synaptic plasticity and adaptive behavior, they still offer valuable insights into how noise and learning mechanisms may influence neural dynamics in computational models. This work lays the foundation for future research aimed at improving both the fidelity of neural simulations and the accuracy of data acquisition techniques.