Directed Functional Connectivity by Variational Cross-mapping of Psychophysiological Variables
Understanding the functional connectivity between different brain regions is vital for improving our comprehension of neural processing and cognition. While directed functional connectivity methods can provide us with statistical estimates of information exchange between regions, classic exploratory methods may not capture the nonlinear temporal effects that are observed in fMRI-BOLD data during task-evoked neural activity. To address this limitation, we propose a novel methodology that leverages variational cross-mapping analysis, inspired by psychophysiological interactions, to identify directional influence between connected regions of interest. Our approach can help uncover previously unknown patterns of information exchange and account for nonlinear effects, making it a valuable addition to the toolkit of researchers studying brain function. We demonstrate the effectiveness of our method using simulated neurovascular signals and publicly available fMRI data from 680 human participants performing an emotional face processing task. Our results suggest information flows from the occipital face area to the superior temporal sulcus and the fusiform face area, and additionally from the superior temporal sulcus to the fusiform gyrus. These findings are consistent with previously documented effective connectivity findings in face processing and provide new insights into the exploratory analyses of non-linear directed connectivity for task-evoked data. Overall, our findings contribute to advancing our understanding of directed functional connectivity in the brain and demonstrate the potential of our method to uncover previously unknown patterns of information exchange. Author summaryThe advent of large datasets has made it possible for many research groups to explore functional connectivity between different brain regions. The ability to assess directed connectivity between multiple regions from task-evoked neural responses could potentially uncover connections that were not previously hypothesized based on available data. However, classic methods for exploring task-evoked effects often rely on specific assumptions that are frequently violated by the data, such as nonlinearity, stationarity, and separability of cause from effect. Recent studies have attempted to address these issues using sliding window approaches or parameterized forward causal models, but these methods have limitations such as fixed contextual effect windows or restricted search space for forward models. To overcome these challenges, we propose a Bayesian non-parametric cross-mapping method that can address non-linearity and separability while using specially designed covariance functions to address non-stationarity. We demonstrate through simulations that our proposed method can detect pair-wise interacting neural populations with high sensitivity and specificity, and accurately infer changes in connections between tasks in both acyclical and cyclical neural networks. We also show that our method can replicate known connectivity findings about emotional face processing in a publicly available dataset. Thus, our method represents a promising exploratory connectivity tool for cognitive and behavioral neurosciences.