Performance of objective functions and optimisation procedures for parameter estimation in system biology models

Abstract

Systems biology: Performance of parameter estimation procedures A systematic comparison of critical choices for faithful parameter-estimation identifies a combination of a hybrid optimisation algorithm (GLSDC) with data-driven normalisation of simulations (DNS) as the generally best option. Experimental data are often provided in relative, arbitrary units. To match simulations to data, two approaches are common: i) using scaling-factors that have to be estimated (SF); or ii) normalising the simulations in the same way as the data (DNS). Using three test-models of increasing complexity, we explored how this choice affects parameter identifiability and estimation performance. We show that in contrast to SF, DNS does not aggravate non-identifiability and a global-hybrid method combined with DNS outperformed local-multi-start methods. The advantage of DNS in terms of estimation speed was particularly pronounced for the most complex test-problem.