Using the pareto frontier to detect deficiencies in a biological simulation model

We introduce the use of multiobjective optimization methods to assess mechanistic simulation models. Such models are often developed to investigate processes underlying phenomena in biology and other fields. The proposed model structure must be assessed to reveal inadequacies in its simulation of the phenomenon. Objective functions are defined to measure how well the model reproduces specific phenomenon features. The objectives may be continuous or binary-valued, e.g. constraints, depending on the quality and quantity of phenomenon data. Assessment requires estimating and exploring the model's Pareto frontier. The problem is illustrated with the assessment of a model of shoot growth in pine trees using an elitist multiobjective evolutionary algorithm (MOEA). The algorithm uses the partition induced on the parameter space by binary-valued objectives. The assessment revealed the need for hysteresis in the model structure to more accurately simulate shoot growth.