Minimax entropy: Optimal statistical models for neural activity and chromatin structure - Project Summary/Abstract Biological functions and structures are inherently emergent, arising from vast networks of interactions at scales below. Clear examples are neural activity and chromatin structure, which, while differing in many respects, both arise from intricate webs of interactions between neurons or genomic loci. In recent years, the study of these and other complex living systems has been revolutionized by experimental advances on unprecedented scales. However, a complete understanding also requires quantitative models that are capable of bridging the gap between large-scale phenomena and fine-scale interactions. The primary roadblock in constructing these methods is the exponential explosion of possible interactions, a problem that becomes even more challenging as experiments grow. This leads to a clear question: Given mea- surements from large-scale experiments, can we infer the most important interactions within a system? Combining ideas from information theory, network science, and statistical physics, my research group has recently translated this question into a precise optimization problem. However, solving this problem in large-scale data poses fundamental computational challenges. The central goal of my research group— and the focus of this proposal—is to overcome these challenges by developing scalable methods for inferring the most important interactions within large biological networks. In turn, these methods will yield optimized models for predicting collective functions and structures from the underlying interactions. We will apply our framework to investigate two model systems: (i) neural activity and (ii) chromatin structure. Leveraging recordings of thousands of neurons in the mouse hippocampus and visual system, we will identify the networks of neural interactions that provide the best description of population-wide activity. Building upon advances in single-cell imaging and chromosome confirmation capture techniques (such as Hi-C), we will infer the interactions between genomic locations that maximally constrain the 3D or- ganization of DNA. In both contexts, our preliminary results indicate that only a small number of important interactions have an outsized effect on the system as a whole. By quantitatively testing this hypothesis in large-scale data, our research will provide critical insights into how populations of neurons encode information and how the genome folds into compact structures that are critical for cellular functions. Together, the proposed work will provide the theoretical and computational tools needed to extend principled statistical models to large living systems. This will prove indispensable in the study of neural activity, chromatin structure, and other collective phenomena that directly impact human health.
