A General Framework for Characterizing Optimal Communication in Brain Networks

Reviewer #1 (Public review):

Summary:

In this study, Fakhar et al. use a game-theoretical framework to model interregional communication in the brain. They perform virtual lesioning using MSA to obtain a representation of the influence each node exerts on every other node, and then compare the optimal influence profiles of nodes across different communication models. Their results indicate that cortical regions within the brain's "rich club" are most influential.

Strengths:

Overall, the manuscript is well-written. Illustrative examples help to give the reader intuition for the approach and its implementation in this context. The analyses appear to be rigorously performed and appropriate null models are included.

Weaknesses:

The use of game theory to model brain dynamics relies on the assumption that brain regions are similar to agents optimizing their influence, and implies competition between regions. The model can be neatly formalized, but is there biological evidence that the brain optimizes signaling in this way? This could be explored further. Specifically, it would be beneficial if the authors could clarify what the agents (brain regions) are optimizing for at the level of neurobiology - is there evidence for a relationship between regional influence and metabolic demands? Identifying a neurobiological correlate at the same scale at which the authors are modeling neural dynamics would be most compelling.

It is not entirely clear what Figure 6 is meant to contribute to the paper's main findings on communication. The transition to describing this Figure in line 317 is rather abrupt. The authors could more explicitly link these results to earlier analyses to make the rationale for this figure clearer. What motivated the authors' investigation into the persistence of the signal influence across steps?

The authors used resting-state fMRI data to generate functional connectivity matrices, which they used to inform their model of neural dynamics. If I understand correctly, their functional connectivity matrices represent correlations in neural activity across an entire fMRI scan computed for each individual and then averaged across individuals. This approach seems limited in its ability to capture neural dynamics across time. Modeling time series data or using a sliding window FC approach to capture changes across time might make more sense as a means of informing neural dynamics.

The authors evaluated their model using three different structural connectomes: one inferred from diffusion spectrum imaging in humans, one inferred from anterograde tract tracing in mice, and one inferred from retrograde tract-tracing in macaque. While the human connectome is presumably an undirected network, the mouse and macaque connectomes are directed. What bearing does experimentally inferred knowledge of directionality have on the derivation of optimal influence and its interpretation?

It would be useful if the authors could assess the performance of the model for other datasets. Does the model reflect changes during task engagement or in disease states in which relative nodal influence would be expected to change? The model assumes optimality, but this assumption might be violated in disease states.

The MSA approach is highly computationally intensive, which the authors touch on in the Discussion section. Would it be feasible to extend this approach to task or disease conditions, which might necessitate modeling multiple states or time points, or could adaptations be made that would make this possible?

Reviewer #2 (Public review):

Summary:

The authors provide a compelling method for characterizing communication within brain networks. The study engages important, biologically pertinent, concerns related to the balance of dynamics and structure in assessing the focal points of brain communication. The methods are clear and seem broadly applicable, however further clarity on this front is required.

Strengths:

The study is well-developed, providing an overall clear exposition of relevant methods, as well as in-depth validation of the key network structural and dynamical assumptions. The questions and concerns raised in reading the text were always answered in time, with straightforward figures and supplemental materials.

Weaknesses:

The narrative structure of the work at times conflicts with the interpretability. Specifically, in the current draft, the model details are discussed and validated in succession, leading to confusion. Introducing a "base model" and "core datasets" needed for this type of analysis would greatly benefit the interpretability of the manuscript, as well as its impact.