Abstract

Disordered semiflexible polymer networks are essential structural components of biological cells and tissues. The mechanical properties of these networks are often highly strain-dependent and can vary significantly depending on subtle details of their underlying structures, which are especially dynamic in living cells. Typically, probing the mechanical response of specific network structures requires computationally expensive simulations. Here, we explore geometric deep learning methods for predicting mechanical behavior directly from structural information. We demonstrate that equivariant graph neural networks can learn to robustly predict key features of the linear and nonlinear viscoelasticity of disordered polymer networks from undeformed network configurations. Then, we show how this approach can be extended to enable the design of networks with tailored mechanical properties.