Gaussian Process-based Inference toward Revealing Brain Functional Connectivity

In the field of neuroscience, the task of accurately deciphering brain connectivity from observed data has continued to receive increased attention. In this paper, we address the challenge of inferring candidates for brain functional connectivity using local field potential data, taking into account nonlinear interactions and multiple delays. Our approach leverages Gaussian processes and automatic relevance determination kernels to learn mapping functions from one brain area to another. The resulting learned topology is represented as a directed graph with an adjacency matrix. We validate the approach on both synthetic computational neural datasets and real macaque datasets. The results demonstrate the capability of the method to successfully reveal synthetic graph structures and uncover biologically meaningful pathways in real-world data.