Model-based QTL detection is sensitive to slight modifications in model formulation

Abstract

Classical crop models have been developed to predict crop yield and quality, and they are based on physiological and environmental inputs. After molecular discoveries, models should integrate genetic variation to allow predictions that are more genotype-dependent. An interesting approach, Quantitative Trait Locus (QTL)-based ecophysiological modeling, has shown promising results for the design of ideotypes that are adapted to biotic and abiotic stresses, but there are still limitations to attaining a fully integrated model. The aim of this case study is to clarify the impact of choosing different model equations (closely related and with different numbers of parameters) and optimization methods on the detection of QTLs controlling the parameters of crop growth. Different growth equations were parameterized based on a genetic population by following different approaches. The correlations between parameters were analyzed, and two different strategies were adopted to address the correlation issue. QTL analysis was performed on the optimized values of the parameters of the growth equations and on the observed dry mass (DM) data to validate the QTLs detected. Overall, models and strategies resulted in different QTLs being detected. Similar LOD profiles but with peaks of different heights were observed, some of which were significant, resulting in different numbers of QTLs. In some cases, peaks had slightly different positions or were absent. Even closely related growth models led to the detection of different QTLs. The goodness of fit and complexity of the growth models were found to be insufficient to select the best model. Calculating parameters independently of observed data may not be a good strategy, whereas setting parameters independent of the genotype is recommended. Given the large-scale global optimization problem and the strong correlations between parameters, the two algorithms tested showed poor performance. Currently, the lack of effective algorithms is the main obstacle to answering the question posed. The authors therefore suggest testing different model formulations and comparing the QTLs detected before choosing the best formulation to use in an ecophysiological modeling approach based on QTLs.

The potential of crop models in simulation of barley quality traits under changing climates: A review

“…However, most of the quality simulation modules developed for crop modeling platforms do not consider the variability of genotypes in quality traits using fixed parameters due to the lack of data and knowledge on complex interactions among different drivers of grain quality (Bertin et al, 2010). One possibility of estimating a suitable parameter range for grain quality traits simulations is linking them to QTLs or genes (Barrasso et al, 2019). Such gene base modules coupled to crop models can also translate gene-by-gene (epistatic) and gene-by-environment interactions on grain quality under different climate change and adaptation scenarios.…”

In silico mapping in an oil palm breeding program reveals a quantitative and complex genetic resistance to Ganoderma boninense

“…We subsequently used these GCA values in FlexQTL because this software requires only one value per genotyped individual whereas they were progeny tested in the pre-nursery trials. Such a two-stage approach could affect QTL results so one-stage approaches are preferred when possible (Xue et al 2017;Barrasso et al 2019). The two types of mixed model used in this study did not lead to major differences in the QTLs identified, and a one-stage IBDbased variance component approach previously reported for production traits (IBD-VC, Tisné et al 2015) that we used on pre-nursery Ganoderma data also produced similar results (data not shown).…”

Reducing a model of sugar metabolism in peach to catch different patterns among genotypes

“…Given the high correlation among parameters describing fruit growth curves [36], model inputs, such as f ruit weight, were randomly assigned using a uniform distribution picking one of the observed growth dynamics and adding an overall random variation between zero and 10% on fruit weight. Finally, shifts in the duration of fruit development among genotypes were also considered.…”