A Gradient Descent Method for Optimization of Model Microvascular Networks

TL;DR

This paper introduces a gradient descent method for optimizing microvascular networks based on transport functions, enabling analysis of their organization and comparison with damaged networks, with applications to understanding biological transport principles.

Contribution

The authors develop a novel gradient descent algorithm that incorporates Kirchhoff's laws for optimizing transport networks, including microvascular blood flow, with potential to analyze various biological network configurations.

Findings

Validated the hypothesis that microvascular networks optimize uniform RBC flow partitioning.

The method can incorporate blood rheology effects like the Fahraeus-Lindqvist effect.

Framework allows exploration of tradeoffs between different network optimization objectives.

Abstract

Within animals, oxygen exchange occurs within networks containing potentially billions of microvessels that are distributed throughout the animal's body. Innovative imaging methods now allow for mapping of the architecture and blood flows within real microvascular networks. However, these data streams have so far yielded little new understanding of the physical principles that underlie the organization of microvascular networks, which could allow healthy networks to be quantitatively compared with networks that have been damaged, e.g. due to diabetes. A natural mathematical starting point for understanding network organization is to construct networks that are optimized accordingly to specified functions. Here we present a method for deriving transport networks that optimize general functions involving the fluxes and conductances within the network. In our method Kirchoff's laws are imposed via Lagrange multipliers, creating a large, but sparse system of auxiliary equations. By treating network conductances as adiabatic variables, we derive a gradient descent method in which conductances are iteratively adjusted, and auxiliary variables are solved for by two inversions of O(N^2) sized sparse matrices. In particular our algorithm allows us to validate the hypothesis that microvascular networks are organized to uniformly partition the flow of red blood cells through vessels. The theoretical framework can also be used to consider more general sets of objective functions and constraints within transport networks, including incorporating the non-Newtonian rheology of blood (i.e. the Fahraeus-Lindqvist effect). More generally by forming linear combinations of objective functions, we can explore tradeoffs between different optimization functions, giving more insight into the diversity of biological transport networks seen in nature.