Identifying branching principles in biological networks using imaging, modeling, and machine learning

Branching in vascular networks and in overall organismic form is one of the most common and ancient features of multicellular plants, fungi and animals. These networks deliver resources and eliminate wastes from early development onward, and even play a vital role in the growth, prognosis, and treatment of tumors and stroke recovery. Because of these basic and applied reasons there is immense interest in identifying key features of vascular branching and their connection to biological function. Here we classify diverse branching networks-mouse lung, human head and torso, angiosperm plants, and gymnosperm plants-by harnessing recent advances in medical imaging, algorithms and software for extracting vascular data, theory for resource-distribution networks, and machine-learning. Specifically, we apply standard machine-learning techniques to a variety of feature spaces. Our results show that our theoretically-informed feature spaces-especially those that determine blood flow rate-combined with Kernel Density Estimation are best at distinguishing networks. Our categorization of networks enhances the mapping between biologic function-such as the dependence of metabolic rate on body mass-to vascular branching traits among organisms and organs. We accomplish this by analyzing how variation in metabolic scaling exponents-around the canonical value of 3/4-arises despite differences in vascular traits. Our results reveal how network categorization and variation in metabolic scaling are both heavily determined by scaling ratios of vessel radii-changes and asymmetries across branching generations-that strongly constrain rates of fluid flow. These linkages will improve understanding of evolutionary convergence across plants and animals while also potentially aiding prognosis and treatment of vascular pathologies and other diseased states.