Modelling Alzheimer's Protein Dynamics: A Data-Driven Integration of Stochastic Methods, Machine Learning and Connectome Insights

Alec MacIver,Hina Shaheen
2024-11-05
Abstract:Alzheimer's disease (AD) is a complex neurodegenerative disorder characterized by the progressive accumulation of misfolded proteins, leading to cognitive decline. This study presents a novel stochastic modelling approach to simulate the propagation of these proteins within the brain. We employ a network diffusion model utilizing the Laplacian matrix derived from MRI data provided by the Human Connectome Project (<a class="link-external link-https" href="https://braingraph.org/cms/" rel="external noopener nofollow">this https URL</a>). The deterministic model is extended by incorporating stochastic differential equations (SDEs) to account for inherent uncertainties in disease progression. Introducing stochastic components into the model allows for a more realistic simulation of the disease due to the multi-factorial nature of AD. By simulation, the model captures the variability in misfolded protein concentration across brain regions over time. Bayesian inference is a statistical method that uses prior beliefs and given data to model a posterior distribution for relevant parameter values. This allows us to better understand the impact of noise and external factors on AD progression. Deterministic results suggest that AD progresses at different speeds within each lobe of the brain, moreover, the frontal takes the longest to reach a perfect disease state. We find that in the presence of noise, the model never reaches a perfect disease state and the later years of AD are more unpredictable than earlier on in the disease. These results highlight the importance of integrating stochastic elements into deterministic models to achieve more realistic simulations, providing valuable insights for future studies on the dynamics of neurodegenerative diseases.
Neurons and Cognition
What problem does this paper attempt to address?
This paper aims to solve the problem of dynamic modeling of misfolded protein propagation in the brain in Alzheimer's disease (AD). Specifically, the researchers proposed a new modeling method that combines stochastic methods, machine learning, and connectomics insights to more realistically simulate the propagation process of misfolded proteins in AD. By introducing stochastic differential equations (SDEs), the model can better reflect the uncertainty in disease progression, thereby providing more realistic disease simulation results. In addition, the study also used the Bayesian inference method to estimate the posterior distribution of relevant parameter values, further improving the predictive ability and robustness of the model. ### Main Objectives of the Paper: 1. **Establish a more realistic AD protein dynamics model**: Improve the simulation accuracy of AD progression by integrating the deterministic network diffusion model and stochastic elements. 2. **Evaluate the impact of noise on AD progression**: Simulate the impact of external factors on AD progression through white noise (Wiener Process) of different intensities, and explore how noise changes the prediction results of the model. 3. **Use the Bayesian inference method for parameter estimation**: Infer the posterior distribution of relevant parameters from the simulation data through the approximate Bayesian computation (ABC) method to verify the validity and reliability of the model. ### Main Findings: - **Results of the deterministic model**: The progression speed of AD in different regions of the brain is different, and the frontal lobe is the region that takes the longest to reach the full - disease state. Most nodes reach a 50% misfolded protein concentration in about 20 years, and after 35 years, the brain is almost completely occupied by misfolded proteins. - **Results of the stochastic model**: After introducing noise, the model shows that under high - noise intensity (σ = 2), the variation in misfolded protein concentration is greater, especially in the later stage after 25 years. The frontal lobe shows a more obvious delay under high - noise conditions. - **Results of Bayesian inference**: The posterior distribution shows that in the presence of noise, the progression of AD has a high degree of uncertainty, especially in the late stage. The average concentration in the frontal lobe usually does not reach 1 at 30 years, indicating that the frontal lobe is more sensitive to noise. ### Conclusion: This study emphasizes the importance of considering noise in multi - factor disease modeling. By introducing stochastic elements, the model can more realistically reflect the complexity and uncertainty of AD, providing valuable insights for future research on neurodegenerative diseases. These results not only help to understand the propagation mechanism of AD but may also have an important impact on the understanding of treatment strategies and disease resistance.