Resource Usage Prediction for Parameter Sweeps with Biochemical System Simulations

Sammanfattning: Exploring the behavior of biochemical systems when subjected to certain internal and external changes is fascinating, and these variations can be investigated through computational simulations. However, the computational cost of simulations is often quite high, necessitating an understanding of the computational requirements and resource utilization of these simulations. In this project, our focus is on studying the time cost of simulations. The objective of this project is to establish a model for predicting the runtime cost of running biochemical simulations.  The Predator-Prey model is a biochemical model used for the analysis. First, parameter sweeps are performed in the parameter spaces to simulate and record the corresponding running times, followed by an analysis of the parameter sweeps dataset. Subsequently, based on parameter combinations and runtimes, a batch K-NN model is established to demonstrate the feasibility of using an online model, and then an online K-NN model is developed to predict the time cost of Predator-Prey model simulations. We find that the online K-NN model performs well for datasets without timeout records, but it fails to handle datasets with multiple timeouts. This predictive model is valuable in estimating simulation time costs and significantly improves upon the prediction method that considers all runtimes as timeouts.