Modelling Alzheimer’s from Amyloid to Tau

AI can be used to model the spread of Alzheimer’s disease through the brain and has now provided researchers with a more biologically grounded way to predict cognitive decline. Details are reported in the International Journal of Simulation and Process Modelling. The work takes into account a shift in neuroscience that now seeks to treat dementia as a dynamic network disorder rather than a static accumulation of toxic proteins.

Nevertheless, the research focuses on Tau, a protein increasingly seen as central to the progression of Alzheimer’s disease. Although the condition is also associated with amyloid plaques, scientists now believe Tau pathology correlates more directly with neurone death and the deterioration of memory and reasoning. Amyloid plaques are perhaps the trigger, but the accumulation of misfolded Tau proteins, which multiply like prions, is thought to be the abnormality that leads to the cognitive problems seen in Alzheimer’s disease.

The new model, NSTP-Net, combines two forms of AI. One is a graph neural network, a type of deep learning designed to analyse interconnected systems. In this case, the brain is represented as a network of linked regions, enabling the model to simulate how disease-related signals travel across neural pathways. The second component uses symbolic reasoning, in which established biological knowledge is encoded directly into the system as logical rules. These include the tendency of Tau to spread along synaptic connections, the vulnerability of highly active brain regions, and the role of genetic risk factors.

The researchers validated their model against data from 428 participants in the Alzheimer’s Disease Neuroimaging Initiative. NSTP-Net was able to reduce prediction error by about 22 per cent compared with existing methods when forecasting Tau spread over an 18-month period. It also showed strong performance in predicting which patients with mild cognitive impairment, measurable memory problems not yet severe enough to qualify as dementia, would later progress to Alzheimer’s disease.

Huo, M., Chen, Y. and Wang, H. (2026) ‘Tau protein transmission simulation modelling in Alzheimer’s disease integrated with neuro-symbolic learning’, Int. J. Simulation and Process Modelling, Vol. 23, No. 6, pp.1–12.