Function with Fluctuations

Many proteins exhibit cooperative binding, in which the binding of a molecule at one site facilitates binding of another molecule at a second site, and vice versa. We show that a symmetric application of Coupled Learning, a local learning rule, can train central spring networks to display cooperative binding. The trained networks have a low-energy vibrational mode that reflects cooperative binding, and that is susceptible to thermal fluctuations. We develop an analytical formulation and use molecular dynamics simulations to verify the breakdown of cooperative binding above a crossover temperature. We find that there is an optimal strain ratio that optimizes thermal resilience and show that the crossover temperature can be raised further by modifying the local learning rule.