Lazy Updating increases the speed of stochastic simulations

Abstract

Biological reaction networks often contain what might be called 'hub molecules', which are involved in many reactions. For example, ATP is commonly consumed and produced. When reaction networks contain molecules like ATP, they are difficult to efficiently simulate, because every time such a molecule is consumed or produced, the propensities of numerous reactions need to be updated. In order to increase the speed of simulations, we developed 'Lazy Updating', which postpones some propensity updates until some aspect of the state of the system changes by more than a defined threshold. Lazy Updating works with several existing stochastic simulation algorithms, including Gillespie's direct method and the Next Reaction Method. We tested Lazy Updating on two example models, and for the larger model it increased the speed of simulations over eight-fold while maintaining a high level of accuracy. These increases in speed will be larger for models with more widely connected hub molecules. Thus Lazy Updating can contribute towards making models with a limited computing time budget more realistic by including previously neglected hub molecules.