Model-based whole-brain perturbational landscape of neurodegenerative diseases
Curation statements for this article:-
Curated by eLife
eLife assessment
Sanz Perl and colleagues provide important insights regarding the application of computational brain models from neurodegenerative diseases to evaluate brain stimulation protocols in silico. Solid evidence is provided for the disease-specificity of the framework, however, the real-world impact of such stimulation protocols to alleviate psychiatric and neurological symptoms remains to be evaluated.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
The treatment of neurodegenerative diseases is hindered by lack of interventions capable of steering multimodal whole-brain dynamics towards patterns indicative of preserved brain health. To address this problem, we combined deep learning with a model capable of reproducing whole-brain functional connectivity in patients diagnosed with Alzheimer’s disease (AD) and behavioral variant frontotemporal dementia (bvFTD). These models included disease-specific atrophy maps as priors to modulate local parameters, revealing increased stability of hippocampal and insular dynamics as signatures of brain atrophy in AD and bvFTD, respectively. Using variational autoencoders, we visualized different pathologies and their severity as the evolution of trajectories in a low-dimensional latent space. Finally, we perturbed the model to reveal key AD- and bvFTD-specific regions to induce transitions from pathological to healthy brain states. Overall, we obtained novel insights on disease progression and control by means of external stimulation, while identifying dynamical mechanisms that underlie functional alterations in neurodegeneration.
Article activity feed
-
-
eLife assessment
Sanz Perl and colleagues provide important insights regarding the application of computational brain models from neurodegenerative diseases to evaluate brain stimulation protocols in silico. Solid evidence is provided for the disease-specificity of the framework, however, the real-world impact of such stimulation protocols to alleviate psychiatric and neurological symptoms remains to be evaluated.
-
Reviewer #1 (Public Review):
In this article, Sanz Perl and colleagues set out to use a computational whole-brain model to simulate the patterns of functional connectivity (as observed from functional MRI) that characterise different forms of dementia, namely Alzheimer's Disease (AD) and behavioural variant frontotemporal dementia (bvFTD). To overall goal is to develop a paradigm to model a specific disorder, and then develop an in silico assessment of the effects of different interventions. They show that superior fitting of the simulated data to the empirical data of both pathologies can be achieved when a Hopf model of brain activity is informed by patterns of combined AD and bvFTD atrophy, or by the intrinsic organisation of brain regions into canonical resting-state networks. They also show that regional differences in the fitted …
Reviewer #1 (Public Review):
In this article, Sanz Perl and colleagues set out to use a computational whole-brain model to simulate the patterns of functional connectivity (as observed from functional MRI) that characterise different forms of dementia, namely Alzheimer's Disease (AD) and behavioural variant frontotemporal dementia (bvFTD). To overall goal is to develop a paradigm to model a specific disorder, and then develop an in silico assessment of the effects of different interventions. They show that superior fitting of the simulated data to the empirical data of both pathologies can be achieved when a Hopf model of brain activity is informed by patterns of combined AD and bvFTD atrophy, or by the intrinsic organisation of brain regions into canonical resting-state networks. They also show that regional differences in the fitted parameters pertain to AD and bvFTD, both in terms of location, and in terms of dynamical regime. They then use a machine learning algorithm, the variational auto-encoder (VAE), to compress functional connectivity patterns into a 2-dimensional space (given by the relative activation of the VAE's two hidden neurons). This space reveals that AD and bvFTD perturb brain connectivity along two distinct dimensions, further stratifying sub-categories of AD. Finally, through visualisation in this latent space, the authors can assess the effects of different simulated interventions on the models previously fitted to AD and bvFTD: namely, stimulation of different regions and with different dynamical regimes, to evaluate whether the resulting model is moved closer to the region occupied by healthy controls.
A strength of this work is its creative combination of different modelling approaches, combining the more biologically-informed Hopf model, which incorporates atrophy maps and connectivity, with the VAE for the purpose of dimensionality reduction and visualisation. Another strength is the use of different controls, such as an atrophy map from a different disorder (Parkinson's) or the use of randomised heterogeneities, showing that the improved fit is not just due to increased degrees of freedom: an important concern for high-dimensional models, which the authors lay to rest.
Admittedly, the stimulation paradigm shows limited success at bringing the disorder-fitted models back to the region occupied by controls - except for the AD- sub-category, which is the least affected and shows the most promise in the authors' in-silico trial. The limited success of this approach in this specific context does not invalidate the framework's promise. This may also be attributed to the fact that the authors do not use disease-specific atrophy maps to model AD and bvFTD: rather, they use a single atrophy map obtained by combining the two and use this joint atrophy map both to model AD, and to model bvFTD. Likewise, the connectivity of the model is the same for all conditions.
A weakness of this work is that, as the authors themselves acknowledge, the brain regions whose stimulation pushes the model to be least far from controls in the latent space did not match with those presenting different bifurcation parameters. In fact, it is not clear whether this is because stimulation fails to reverse the regional alterations of the dynamical regime, or whether it does succeed, but introduces new alterations - although it should be possible to extract this information from the model, to provide additional insight. This raises the intriguing question of the biological meaning of the latent space. Although the authors do show what kinds of FC correspond to the different values of the VAE hidden neurons' activation, the latent space effectively acts as a 2-dimensional goodness-of-fit - raising the question of how much of the stimulation results could be captured by simply evaluating the stimulated model's GOF against controls (while acknowledging that this would conflate the two distinct dimensions along which AD and bvFTD differ from controls).
Since stimulation is intended to mimic the effects of different real-life interventions such as tACS and tDCS, it would be helpful to see whether the regions that are suggested as most promising for stimulation, do in fact match the regions that have shown the most success in actual clinical trials that have already been carried out. This would be a powerful validation from model to real applicability.
In its essence, the work makes progress towards the authors' goal of modelling different pathologies by incorporating biologically-derived information, highlighting their differences, and enabling the evaluation of different stimulation strategies. This computational framework is widely applicable to a variety of pathological (and even non-pathological) conditions, combining evaluation and intervention in a single workflow.
-
Reviewer #2 (Public Review):
The authors present an interesting study combining deep learning, neuroimaging, and brain stimulation techniques for several neurodegenerative diseases. This has important consequences to understand the connectivity alterations and to design novel therapies to alleviate these changes.
-
