Ruben Sanchez-Romero

Functional connectivity (FC) has been invaluable for understanding the brain’s communication network, with strong potential for enhanced FC approaches to yield additional insights. Unlike with the fMRI field-standard method of pairwise correlation, theory suggests that partial correlation can estimate FC without confounded and indirect connections. However, partial correlation FC can also display low repeat reliability, impairing the accuracy of individual estimates. We hypothesized that reliability would be increased by adding regularization, which can reduce overfitting to noise in regression-based approaches like partial correlation. We therefore tested several regularized alternatives—graphical lasso, graphical ridge, and principal component regression—against unregularized partial and pairwise correlation, applying them to empirical resting-state fMRI and simulated data. As hypothesized, regularization vastly improved reliability, quantified using between-session similarity and intraclass correlation. This enhanced reliability then granted substantially more accurate individual FC estimates when validated against structural connectivity (empirical data) and ground truth networks (simulations). Graphical lasso showed especially high accuracy among regularized approaches, seemingly by maintaining more valid underlying network structures. We additionally found graphical lasso to be robust to noise levels, data quantity, and subject motion—common fMRI error sources. Lastly, we demonstrated that resting-state graphical lasso FC can effectively predict fMRI task activations and individual differences in behavior, further establishing its reliability, external validity, and ability to characterize task-related functionality. We recommend graphical lasso or similar regularized methods for calculating FC, as they can yield more valid estimates of unconfounded connectivity than field-standard pairwise correlation, while overcoming the poor reliability of unregularized partial correlation.

Research into the mechanisms underlying neuromodulation by tES using in-vivo animal models is key to overcoming experimental limitations in humans and essential to building a detailed understanding of the in-vivo consequences of tES. Insights from such animal models are needed to develop targeted and effective therapeutic applications of non-invasive brain stimulation in humans. The sheer difference in scale and geometry between animal models and the human brain contributes to the complexity of designing and interpreting animal studies. Here, we introduce EFMouse, a toolbox that extends previous approaches to model intracranial electric fields and generate predictions that can be tested with in-vivo recordings in mice. Novel functionality includes the ability to capture typical surgical approaches in the mouse (e.g., cranial recording windows), the placement of stimulation electrodes anywhere in or on the animal, and novel ways to report field predictions, including some refined measures of focality and direction homogeneity, and quantification based on regions defined in the Allen Mouse Brain Atlas. Although the EFMouse toolbox is generally applicable to planning and designing tES studies in mice, we illustrate its use by posing questions about transcranial direct current stimulation (tDCS) experiments with the goal of targeting the left visual cortex of the mouse. The EFMouse toolbox is publicly available at https://github.com/klabhub/EFMouse.

Compared to the rapidly growing literature on transcranial electrical stimulation (tES) in humans, research into the mechanisms underlying neuromodulation by tES using in-vivo animal models is growing but still relatively rare. Such research, however, is key to overcoming experimental limitations in humans and essential to build a detailed understanding of the in-vivo consequences of tES that can ultimately lead to development of targeted and effective therapeutic applications of noninvasive brain stimulation. The sheer difference in scale and geometry between animal models and the human brain contributes to the complexity of designing and interpreting animal studies. Here we extend previous approaches to model intracranial electric fields to generate predictions that can be tested with in-vivo intracranial recordings. Although the toolbox has general applicability and could be used to predict intracranial fields for any tES study using mice, we illustrate its usage by comparing fields in a high-density multi-electrode montage with a more traditional two electrode montage. Our simulations show that both montages can produce strong focal homogeneous electric fields in targeted areas. However, the high-density montage produces a field that is more perpendicular to the visual cortical surface, which is expected to result in larger changes in neuronal excitability. Highlights - EFMouse is a novel, open-source, Matlab-based electric field simulator for the mouse brain. - EFMouse quantifies field focality and homogeneity in regions of the Allen Mouse Brain Atlas. - Focal and strong stimulation can be produced with two or five electrode montages. - A high-density montage with a lumbar return generates fields perpendicular to the cortical surface.

Arousal state is regulated by subcortical neuromodulatory nuclei, such as locus coeruleus, which send wide-reaching projections to cortex. Whether higher-order cortical regions have the capacity to recruit neuromodulatory systems to aid cognition is unclear. Here, we hypothesized that select cortical regions activate the arousal system, which in turn modulates large-scale brain activity, creating a functional circuit predicting cognitive ability. We utilized the Human Connectome Project 7T functional magnetic resonance imaging dataset (N=149), acquired at rest with simultaneous eye tracking, along with extensive cognitive assessment for each subject. First, we discovered select frontoparietal cortical regions that drive large-scale spontaneous brain activity specifically via engaging the arousal system. Second, we show that the functionality of the arousal circuit driven by bilateral posterior cingulate cortex (associated with the default mode network) predicts subjects cognitive abilities. This suggests that a cortical region that is typically associated with self-referential processing supports cognition by regulating the arousal system.

Learning graphical conditional independence structures is an important machine learning problem and a cornerstone of causal discovery. However, the accuracy and execution time of learning algorithms generally struggle to scale to problems with hundreds of highly connected variables -- for instance, recovering brain networks from fMRI data. We introduce the best order score search (BOSS) and grow-shrink trees (GSTs) for learning directed acyclic graphs (DAGs) in this paradigm. BOSS greedily searches over permutations of variables, using GSTs to construct and score DAGs from permutations. GSTs efficiently cache scores to eliminate redundant calculations. BOSS achieves state-of-the-art performance in accuracy and execution time, comparing favorably to a variety of combinatorial and gradient-based learning algorithms under a broad range of conditions. To demonstrate its practicality, we apply BOSS to two sets of resting-state fMRI data: simulated data with pseudo-empirical noise distributions derived from randomized empirical fMRI cortical signals and clinical data from 3T fMRI scans processed into cortical parcels. BOSS is available for use within the TETRAD project which includes Python and R wrappers.

Functional connectivity (FC) has been invaluable for understanding the brain's communication network, with strong potential for enhanced FC approaches to yield additional insights. Unlike with the fMRI field-standard method of pairwise correlation, theory suggests that partial correlation can estimate FC without confounded and indirect connections. However, partial correlation FC can also display low repeat reliability, impairing the accuracy of individual estimates. We hypothesized that reliability would be increased by adding regularization, which can reduce overfitting to noise in regression-based approaches like partial correlation. We therefore tested several regularized alternatives — graphical lasso, graphical ridge, and principal component regression — against unregularized partial and pairwise correlation, applying them to empirical resting-state fMRI and simulated data. As hypothesized, regularization vastly improved reliability, quantified using between-session similarity and intraclass correlation. This enhanced reliability then granted substantially more accurate individual FC estimates when validated against structural connectivity (empirical data) and ground truth networks (simulations). Graphical lasso showed especially high accuracy among regularized approaches, seemingly by maintaining more valid underlying network structures. We additionally found graphical lasso to be robust to noise levels, data quantity, and subject motion — common fMRI error sources. Lastly, we demonstrated that resting-state graphical lasso FC can effectively predict fMRI task activations and individual differences in behavior, further establishing its reliability, external validity, and ability to characterize task-related functionality. We recommend graphical lasso or similar regularized methods for calculating FC, as they can yield more valid estimates of unconfounded connectivity than field-standard pairwise correlation, while overcoming the poor reliability of unregularized partial correlation.

Editorial: Brain connectivity, dynamics, and complexity

In seeking to understand processes fundamental to brain function, the debate between localized versus distributed processing has persisted for over a century. High category selectivity of some visual regions has provided evidence for localized (within-region) processing. However, given recent evidence that distributed connectivity patterns are predictive of localized task-evoked activations, we hypothesized that even highly localized category-selective brain activations are primarily determined by the connectivity-specified convergence of distributed brain processes. Consistent with this, we utilized fMRI data from N=352 human participants to find that distributed activity flow processes (specified by each region’s unique “connectivity fingerprint”) locally converge on visual cortex regions to confer the majority contribution to visual category selective responses to bodies, faces, places, and tools. This highlights a prominent role for functional network organization in the generation of localized visual category selectivity and provides evidence for a distributed convergence account of localized functionality in the human brain. Significance Statement A fundamental debate in neuroscience has persisted for over a century with strong evidence on both sides: to what extent do localized versus distributed processes drive brain function? The existence of category-selective regions within the visual cortex provides some of the clearest evidence supporting localized processing, wherein specialized functions (e.g., category selectivity to faces) are primarily generated by within-region processes. This localized account has been countered by the identification of distributed visual cortex category selectivity in the absence of category-selective regions. Here we provide groundwork evidence for a theory of distributed convergence, demonstrating that localized visual-category-selective responses can be accurately generated via distributed activity flow processes. These predictions were generated via each category-selective region’s unique connectivity fingerprint, demonstrating that the human brain’s distributed network architecture is sufficient to generate category selectivity in visual cortex regions that were historically seen as evidence for a primarily local account.