Trainability in biopolymer-inspired disordered spring networks

Disordered networks are ubiquitous in nature, particularly in biological systems where they serve as the mechanical foundation of structures ranging from proteins to the cytoskeleton. Previously it was shown that disordered spring networks can be tuned to exhibit target mechanical properties by selectively removing individual bonds, but breaking events in biological networks usually involve more complicated changes of the network topology. Inspired by this, we introduce a disordered spring network model that shows the basic features observed in biopolymer networks, including force/strain-induced changes in topology. Nodes can be removed/broken, analogous to crosslink unbinding, with the edges attached to it being redistributed to neighboring nodes while preserving topology and conserving rest length (akin to preserving mass). We simulate these networks under cyclic strain protocols and determine the mechanical properties of the resulting networks after breaking events. The resulting networks are stressed even in the unstrained state due to rearrangement of the edges. The networks studied are trainable to develop different desired mechanical properties when the initial state has been prepared to have suitable disorder and internal stresses.