Discovering Relationships Across Disparate Data Modalities

Discovering whether certain properties are associated with other properties is fundamental to all science. As the amount of data increases, it is becoming increasingly difficult and important to determine whether one property of the data (e.g., cloud density) is related to another (e.g., grass wetness). Only If they are related does it make sense to further investigate the nature of the relationship. Unfortunately, reliably identifying relationships can be challenging, especially when the properties themselves are complex and the relationship is nonlinear and high-dimensional. Here, we describe a procedure, Multiscale Generalized Correlation (MGC), that addresses these challenges. Our key insight is that if two properties are related, comparisons between measurements of similar pairs of the first property (e.g., two clouds of similar density) should be correlated with the comparisons between corresponding measurements of the second property (grass wetness under those clouds). We demonstrate the statistical and computational efficiency of MGC in both simulations and theory. We then apply it to detect the presence and nature of the relationships between brain activity and personality, brain shape and disorder, and brain connectivity and creativity. Finally, we demonstrate that MGC does not suffer from the false positives that have plagued parametric methods. Our open source implementation of MGC is applicable to fundamental questions confronting science, government, finance, and many other disciplines.