Inferring relationships among major psychiatric disorders in a resting-state functional connectivity-informed embedding space

Major neuropsychiatric disorders such as major depressive disorder (MDD) and schizophrenia (SCZ), as well as the neurodevelopmental disorder autism spectrum disorder (ASD), are traditionally treated as distinct clinical entities. However, genome-wide association studies indicate shared genetic risks, motivating a transdiagnostic view. Resting-state functional connectivity (rsFC) is a promising biomarker for these disorders, but its high dimensionality complicates inference of inter-disorder relationships in the native feature space. Here, we develop an rsFC-based embedding-relation workflow that quantifies disorder relationships in a connectivity-informed, low-dimensional embedding space. Central to the workflow is a mutual information-based embedding framework that evaluates candidate embedding approaches and selects an optimal strategy. Using synthetic connectivity data, the framework indicates that rsFC embeddings are best represented in a spherical space under a moderate level of supervision. Building on this insight, we applied the workflow to curated, multi-disorder rsFC datasets to derive shared embedding spaces encompassing the connectivity features of ASD, MDD, and SCZ. In these spaces, we consistently observed a robust three-way relationship: a pronounced neurobiological dissimilarity between ASD and MDD, contrasted with greater similarity between SCZ and both disorders. These findings support a dimensional, transdiagnostic perspective on neuropsychiatric disorders and offer new insights into their shared and distinct neural underpinnings.