Towards unified multimodal cross-scale modelling of neurodegenerative diseases

With ever-increasing life expectancy, neuro- degenerative diseases pose a major global healthcare challenge. Disorders such as Alzheimer’s disease (AD), Parkinson’s disease (PD), and frontotemporal lobar degeneration (FTLD) characteristically follow disease-specific spatiotemporal trajectories. It is now widely recognized that core pathological hallmarks of neurodegeneration—misfolded proteins including hyperphosphorylated tau, α-synuclein, and TAR DNA-binding protein 43—can propagate trans-synaptically across neural circuits [1]. In this context, the brain can be conceptualized as a network system in which disease processes unfold on intrinsic structural and functional connectome. Consistent with this framework, neuroimaging studies have demonstrated close correspondence between the spatial covariance of neurodegenerative pathology and the brain’s underlying connectivity organization [2].

1 Macroscopic spatial modelling of neurodegeneration

Building on the connectome-based disease propagation hypothesis, multiple computational approaches have been proposed to model disease progression. Spectral graph-based models characterize neurodegenerative patterns in terms of eigenmodes of brain networks, capturing dominant axes along which pathology disseminates [3]. A complementary class of models leverages epidemic spreading models (ESMs), in which pathological proteins are simulated as infectious agents that transmit between vulnerable brain regions, with structural or functional connectivity constraining the probability of interregional spread. These models have been shown to reproduce characteristic patterns of pathology accumulation across several neurodegenerative disorders [4].

While the overall progression of a neuro- degenerative disease has been well computed, inter-patient heterogeneity of the disease certainly remains. To tackle this heterogeneity, an alternative perspective conceptualizes neurodegeneration as a sequence of discrete events rather than a continuous diffusion process. Event-based models, particularly the Subtype and Stage Inference (SuStaIn) framework, enable data-driven identification of heterogeneous spatiotemporal trajectories within a disease [5]. By jointly inferring disease subtypes and temporal stages, SuStaIn has revealed substantial heterogeneity within conditions such as genetic frontotemporal dementia, AD, Lewy body dementia, and TDP-43 proteinopathies.

2 Microscopic evidence on neurodegenerative mechanisms

Differential gene expression may be an explanation, also a solution, to the heterogeneity within a neurodegenerative disease, and thus to why and how pathological proteins emerge in specific brain regions. At the microscopic scale, advances in molecular neuroscience have provided critical insights into mechanisms underlying selective neuronal vulnerability and resilience. Spatial and single-cell transcriptomic studies have identified specific clusters of microglia and inflammatory modules contributing to the risk of neurodegeneration, as well as pathways associated with regional vulnerability to specific pathologies [6].

3 Closing the gap: leveraging data across scale

Despite accumulative evidence digging into mechanisms of neurodegenerative diseases at both macroscopic and microscopic levels, a critical challenge remains: how to integrate these complementary perspectives. In other words, how to integrate the existing macroscopic and microscopic knowledge into a unified, mechanistically interpretable model of neuro- degenerative diseases. Practically, large-scale resources, including normalized whole-brain transcriptomic maps from the Allen Human Brain Atlas and population-level connectomes from the Human Connectome Project, have laid the foundation for such effort. Indeed, some initial attempts to bridge scales show promising results. For example, network-level analyses of tau deposition have linked spatial patterns of pathology propagation to regional gene expression gradients, implicating genes such as APOE and SLC1A2 in network-mediated tau spread [7]. Other studies have examined whether transcriptomic profiles can explain residual variance in disease patterns beyond what is accounted for by connectivity-based propagation models [8]. However, most existing approaches remain loosely integrative, underscoring the need to develop a model that is simultaneously multiscale, mechanistically grounded, and clinically actionable.

Beyond providing insights for disease mechanisms, multimodal cross-scale modelling holds significant promises for precision medicine. Personalized disease-forecasting frameworks have demonstrated that individual-specific anatomical patterns, combined with structural and functional connectivity, can be used to predict future trajectories of brain atrophy [9]. More recently, integrative platforms have expanded this approach by jointly modelling clinical, genetic, histopathological, and neuroimaging data, enabling a more comprehensive characterization of disease heterogeneity [10].

4 Potential of unified integrative models

A unified multimodal, cross-scale modelling paradigm—linking molecular biology to network dynamics—has a potential to redefine how neurodegenerative diseases are conceptualized, shifting the field from identifying isolated biomarkers towards modelling disease progression in a mechanistic, patient-specific approach. Specifically, from mechanistic perspective, omics profile may explain regional susceptibility to certain neurodegenerative pathology, and the propensity to propagate along certain network connections [11]. While in individualised prediction models, clinical, transcriptomic and imaging profile could be used to inform tailored intervention strategy to mitigate network progression [12].

Advancing this field will require sustained efforts in high-quality multimodal data acquisition, methodological innovation, and close collaboration across disciplines. Positioned at the intersection of clinical neuroscience and data science, unified multimodal modelling offers a powerful framework to link molecular mechanisms with network- level disease dynamics. As multimodal datasets continue to grow and computational tools to mature, such approaches are poised to generate transformative insights into the mechanisms of neurodegeneration and to inform individualized interventions in clinical practice.