Expert-guided multi-objective optimization: An efficient strategy for parameter estimation of biological systems with limited data

Calibrating biological models is challenging due to high-dimensional parameter spaces and the limited availability of reliable experimental data. In this study, we propose a hybrid calibration framework that integrates expert knowledge into a multi-objective optimization process. We have evaluated three multi-objective optimization algorithm (NSGA-III, MOEA/D and MO-TPE) with our framework to combine hard constraints derived from biological measurements with soft constraints encoding qualitative domain expertise, such as expected curve shapes or event timing. This dual-constraint strategy guides the search toward biologically plausible parameter sets while preserving flexibility and interpretability. We demonstrate the effectiveness of our method on a benchmark model of skin wound healing, comparing it to standard and unconstrained optimization strategies. Results show that the framework reduces the risk of overfitting to sparse time-course data by favoring dynamically plausible trajectories that satisfy expert-guided soft constraints, increasing the proportion of biologically plausible solutions generated from 1.8% ± 1.3 to 24.3% ± 8.6 for NSGA-III without constraint to NSGA-III with 6 constraints, respectively (p<0.0001). The framework is flexible, iterative, and generalizable, offering a principled way to leverage domain knowledge for model calibration in complex biological systems.