Combining multiple functional connectivity methods to improve causal inferences

Abstract

Cognition and behavior emerge from brain network interactions, suggesting that causal interactions should be central to the study of brain function. Yet approaches that characterize relationships among neural time series-functional connectivity (FC) methods-are dominated by methods that assess bivariate statistical associations rather than causal interactions. Such bivariate approaches result in substantial false positives since they do not account for confounders (common causes) among neural populations. A major reason for the dominance of methods such as bivariate Pearson correlation (with functional MRI) and coherence (with electrophysiological methods) may be their simplicity. Thus, we sought to identify an FC method that was both simple and improved causal inferences relative to the most popular methods. We started with partial correlation, showing with neural network simulations that this substantially improves causal inferences relative to bivariate correlation. However, the presence of colliders (common effects) in a network resulted in false positives with partial correlation, though this was not a problem for bivariate correlations. This led us to propose a new combined functional connectivity method (combinedFC) that incorporates simple bivariate and partial correlation FC measures to make more valid causal inferences than either alone. We release a toolbox for implementing this new combinedFC method to facilitate improvement of FC-based causal inferences.

Estimation of Dynamic Bivariate Correlation Using a Weighted Graph Algorithm

“…Focusing on the pairwise bivariate correlations alone may exclude the potential effects of other regions on the pair of brain regions under consideration. There are new directions proposed recently in the neuroscience literature using partial correlations to account for such confounding effects [ 36 , 37 ]. To control for false positives due to partial correlations in the presence of colliders, [ 36 , 37 ] proposed a combined approach that includes both partial correlations and pairwise bivariate correlations.…”

Tracking the Progression & Influence of Beta-Amyloid Plaques Using Percolation Centrality and Collective Influence Algorithm: A Study using PET Images

“…Whereas in cases of “colliding” interactions, a partial correlation may induce a spurious correlation. Thus, Sanchez-Romero and Cole have introduced a combined multiple functional connectivity method[47].…”

Network modeling of dynamic brain interactions predicts emergence of neural information that supports human cognitive behavior

“…Thirdly, our MVAR approach fitted all spatial ("source" predictor regions) and temporal (t0-n lags) terms simultaneously via multiple linear regression models ( Figure 3A), meaning that the approach is fully multivariate in both the temporal and spatial domains. This prevents indirect spatial or temporal influences from serving as unobserved causal confounders on the FC weights, which would arise if alternative bivariate FC estimation methods were used (Pearl and Mackenzie, 2018;Reid et al, 2019;Sanchez-Romero and Cole, 2020). Fourthly, our exclusive focus on the lagged MVAR estimates for dynamic activity flow modeling (see later Method section for details) imparted directionality to the approach, given that past t0-n lagged terms predicted future target t0 terms ( Figure 3B).…”

“…Given that it is impossible for brain activity to propagate back in time (a fundamental principle of causality and the direction of time), this completely eliminates the possibility of analytic circularity. The resultant “dynamic activity flow modeling” approach properly leverages the higher temporal resolution of EEG data to overcome a limitation of the method and, due to the use of lagged multivariate FC (Liégeois et al, 2017; Mill et al, 2017; Sanchez-Romero and Cole, 2020), open up enhanced causal inferences relative to standard models.…”