Recent studies have shown that novel genetic variation for resistance to pests and diseases can be detected in plant genetic resources originating from locations with an environmental profile similar to the collection sites of a reference set of accessions with known resistance, based on the Focused Identification of Germplasm Strategy (FIGS) approach. FIGS combines both the development of a priori information based on the quantification of the trait-environment relationship and the use of this information to define a best bet subset of accessions with a higher probability of containing new variation for the sought after trait(s). The present study investigates the development strategy of the a priori information using different modeling techniques including learning-based techniques as a follow up to previous work where parametric approaches were used to quantify the stem rust resistance and climate variables relationship. The results show that the predictive power, derived from the accuracy parameters and cross-validation, varies depending on whether the models are based on linear or non-linear approaches. The prediction based on learning techniques are relatively higher indicating that the non-linear approaches, in particular support vector machine and neural networks, outperform both principal component logistic regression and generalized partial least squares. Overall there are indications that the trait distribution of resistance to stem rust is confined to certain environments or areas, whereas the susceptible types appear to be limited to other areas with some degree of overlapping of the two classes. The results also point to a number of issues to consider for improving the predictive performance of the models.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below