From descriptive connectome to mechanistic connectome: Generative modeling in functional magnetic resonance imaging analysis

Abstract

As a newly emerging field, connectomics has greatly advanced our understanding of the wiring diagram and organizational features of the human brain. Generative modeling-based connectome analysis, in particular, plays a vital role in deciphering the neural mechanisms of cognitive functions in health and dysfunction in diseases. Here we review the foundation and development of major generative modeling approaches for functional magnetic resonance imaging (fMRI) and survey their applications to cognitive or clinical neuroscience problems. We argue that conventional structural and functional connectivity (FC) analysis alone is not sufficient to reveal the complex circuit interactions underlying observed neuroimaging data and should be supplemented with generative modeling-based effective connectivity and simulation, a fruitful practice that we term “mechanistic connectome.” The transformation from descriptive connectome to mechanistic connectome will open up promising avenues to gain mechanistic insights into the delicate operating principles of the human brain and their potential impairments in diseases, which facilitates the development of effective personalized treatments to curb neurological and psychiatric disorders.

Excitation-inhibition imbalance in Alzheimer’s disease using multiscale neural model inversion of resting-state fMRI

“…Due to the inability of conventional fMRI to resolve excitatory versus inhibitory activities (Devor et al, 2007; Anenberg et al, 2015; Vazquez et al, 2018), a fMRI-based computational framework that could accurately estimate E-I imbalance during AD progression is urgently needed. As the two widely used approaches for generative modeling, DCM (Friston et al, 2003, 2014; Li et al, 2011) and BNM (Honey et al, 2007, 2009; Deco and Jirsa, 2012; Deco et al, 2013a, b) are limited in either the biophysical realism (DCM) or the ability to estimate individual connection strengths (BNM) (see review in Li and Yap, 2022). To overcome these limitations, we applied a recently developed MNMI framework (Li et al, 2019; 2021) to an ADNI dataset to identify disrupted E-I balance in a large network during AD progression.…”

“…In contrast, generative modeling-based fMRI analysis has the capability of both inferring E-I balance at circuit level and simulating the impact of E-I imbalance on network dynamics (Li and Yap, 2022). For example, de Hann et al (2012de Hann et al ( , 2017 developed a large-scale neural mass model to examine the effects of excessive neuronal activity on functional network topology and dynamics.…”

“…Despite the great success of conventional fMRI analysis in characterizing altered network activation and connectivity patterns in AD, they are not able to identify E-I imbalance at circuit level because such statistical tools do not allow for determination of causal relationships between regions, nor do they provide insight into the dynamic meso-scale neuronal relationships that underpin BOLD signal variations. In contrast, generative modeling-based fMRI analysis has the capability of both inferring E-I balance at circuit level and simulating the impact of E-I imbalance on network dynamics (Li and Yap, 2022). For example, de Hann et al (2012, 2017) developed a large-scale neural mass model to examine the effects of excessive neuronal activity on functional network topology and dynamics.…”

“…Moreover, most current analytic approaches for fMRI, including graph theory, seed-based approaches, and independent component analysis (Li et al, 2009; Sporns, 2014; Preti et al, 2017) do not allow for determination of causal relationships between regions, nor do they provide insight into the dynamic meso-scale neuronal relationships that underpin blood-oxygen-level-dependent (BOLD) signal variations, thus unable to identify E-I imbalance at circuit levels. Generative modeling, by comparison, builds on biologically plausible models of neural interactions (Friston et al, 2003; Friston, 2011; Stephan et al, 2015; Li and Yap, 2022) and thus can, in principle, resolve excitatory versus inhibitory neuronal activity. For example, de Hann et al (2012, 2017) developed a large-scale neural mass model to examine the effects of excessive neuronal activity on functional network topology and dynamics.…”

Macroscale coupling between structural and effective connectivity in the mouse brain

“…This is more pronounced in larger scales, when bottom-up models become intractable due to their fast complexity growth and top-down models are the solely alternative. Focusing on the macro scale case, which is the target of this work, the top-down models currently available are of two main types (Li and Yap, 2022; D’Angelo and Jirsa, 2022): brain network model (BNM) (Ritter et al, 2013) and dynamic causal modeling (DCM) (Friston et al, 2003). In both cases, SC is often used to constrain the interactions across brain units.…”