Keywords: m. computational models; x. computational models
Modern genetic research focuses on elucidating protein function and pathways. These involve interactions among various entities that influence gene regulation and expression. The field of Systems Biology seeks to devise experimental and computational approaches to understand how these interactions impact the behavior of the whole. However, systems biology models contain many unknown parameters and assumptions, and to improve their accuracy they must be fit to experimental data. One historically successful approach to fit data is to use evolutionary algorithms. These algorithms begin with randomly-selected parameter sets, which are improved in each generation by ranking and selecting the best parameter sets. These parameter sets are then adapted to experimental data using mathematical strategies like “recombination” and “mutation” to reach a global optimum.
Improved Stochastic Ranking Evolution Strategy (ISRES) is one such evolutionary optimization algorithm, developed by Runarsson and Yao, 2005. It uses stochastic ranking and island hopping to solve nonlinearly constrained optimization problems. It is an approach to fit experimental data to a rule-based model and obtain the best-fit parameter set. We modified this algorithm to make it faster and more accurate. We tested this modified algorithm一ISRES-plus一to verify the improved performance using two models from literature for this: Schmierer et. al, 2008 model for TGF- signaling in HaCat Cells and Manu, et. al, 2009 model for gap gene segmentation in Drosophila Melanogaster. We generated a best-fit parameter set with ISRES and ISRES-plus and verified that ISRES-plus is faster, robust, and more accurate than ISRES.