Mechanisms Underlying Disorders of Consciousness: Bridging Gaps to Move Toward an Integrated Translational Science

. 2021 Jul 8;35(Suppl 1):37–54. doi: 10.1007/s12028-021-01281-6

Box 2 Whole-brain computational models
Whole-brain computational models represent a powerful set of tools to study macroscale mechanistic questions in neuroscience [62–64]. Such models typically combine two fundamental ingredients (Fig. 4a): (1) information about brain network structure (e.g., obtained from diffusion-weighted imaging in humans or invasive tract tracing in animals) and (2) a model of regional neural activity, ranging from Kuramoto or Hopf oscillators to the dynamic mean-field model obtained by mean-field reduction of integrate-and-fire spiking neurons with excitatory and inhibitory populations [65]. The complex interactions of these two key components can give rise to rich and biologically realistic functional dynamics analogous to those observed from fMRI and EEG [65, 66]. Although the required level of neurobiological detail will vary according to the specific question under investigation, the more biologically inspired models (e.g., dynamic mean-field) can also be enriched with further information, such as regional myelination or the regional distribution of specific receptors obtained from positron-emission tomography (PET) [64, 67–70]
Importantly, in silico computational models offer several advantages: their parameters are fully available to inspection and manipulation by the researcher, and they can be perturbed in ways that would not be possible in either humans or animals [67–70]. Computational modeling allows formulation and testing of specific mechanisms, a key feature not provided by other techniques (e.g., neuroimaging). Moreover, the same model can be subjected to different kinds of perturbations to investigate which pharmacological or structural interventions have equivalent results on the model’s function (Fig. 4b), a powerful avenue to interrogate potential similarities between anesthesia and DOC. Finally, the advent of computational models offers the unique promise to develop personalized models from each patient’s multimodal neuroimaging data, and subsequently perform systematic perturbation of the model to evaluate the potential effects of different treatment approaches, with the ultimate goal of informing which therapeutic modalities may be applicable for a patient

Box 2 Whole-brain computational models

Whole-brain computational models represent a powerful set of tools to study macroscale mechanistic questions in neuroscience [62–64]. Such models typically combine two fundamental ingredients (Fig. 4a): (1) information about brain network structure (e.g., obtained from diffusion-weighted imaging in humans or invasive tract tracing in animals) and (2) a model of regional neural activity, ranging from Kuramoto or Hopf oscillators to the dynamic mean-field model obtained by mean-field reduction of integrate-and-fire spiking neurons with excitatory and inhibitory populations [65]. The complex interactions of these two key components can give rise to rich and biologically realistic functional dynamics analogous to those observed from fMRI and EEG [65, 66]. Although the required level of neurobiological detail will vary according to the specific question under investigation, the more biologically inspired models (e.g., dynamic mean-field) can also be enriched with further information, such as regional myelination or the regional distribution of specific receptors obtained from positron-emission tomography (PET) [64, 67–70]

Importantly, in silico computational models offer several advantages: their parameters are fully available to inspection and manipulation by the researcher, and they can be perturbed in ways that would not be possible in either humans or animals [67–70]. Computational modeling allows formulation and testing of specific mechanisms, a key feature not provided by other techniques (e.g., neuroimaging). Moreover, the same model can be subjected to different kinds of perturbations to investigate which pharmacological or structural interventions have equivalent results on the model’s function (Fig. 4b), a powerful avenue to interrogate potential similarities between anesthesia and DOC. Finally, the advent of computational models offers the unique promise to develop personalized models from each patient’s multimodal neuroimaging data, and subsequently perform systematic perturbation of the model to evaluate the potential effects of different treatment approaches, with the ultimate goal of informing which therapeutic modalities may be applicable for a patient