Assessing inter-individual genetic variability in peach sugar metabolism through reliable parameter estimation of a kinetic model

Integrating genetic information into dynamical models is crucial for understanding genotype variation and improving crop performance in the face of current agronomic and ecological constraints. Calibration over a large number of genotypes is a key step in building gene-to-phenotype models but remains challenging. We compared two strategies for calibrating an Ordinary Differential Equations (ODE) kinetic model simulating sugar accumulation during peach fruit development. In the first strategy, a Genotype-Based (GB) approach, the model was calibrated independently for each genotype, using either a Single-Objective Optimization (GBS) or a Multi-Objective Optimization (GBM) formulation. In the second strategy, a Population-Based (PB) approach, the model was calibrated for all genotypes simultaneously. The two strategies were applied to simulated data and to a real dataset from 106 peach genotypes. Results showed that the GB strategy allowed for a high goodness of fit for most genotypes, especially in the GBS formulation. However, the estimated parameters suffered from a lack of practical identifiability as independent repetitions of the estimation algorithm did not always converge to the same value for most genotypes. The PB calibration strategy overcame this issue showing a good identifiability of the population parameter values, a goodness of fit comparable to the one obtained with the GB strategy, and a good characterization of parameter variations within the progeny, which is key to assess the inter-individual genetic variability. This study demonstrates the value of the PB calibration strategy in capturing inter-individual genetic variability, a critical step toward developing robust gene-to-phenotype models. • The calibration of gene-to-phenotype (G2P) models is challenging for plant breeding. • A kinetic model of fruit sugar metabolism was calibrated over 106 peach genotypes. • Genotype-based (GB) and population-based (PB) calibration strategies were compared. • The PB strategy shows better performances compared to the GB strategy. • This is an important step towards the development of reliable G2P models.