Xa gene recommendation was based on gene effectiveness prediction from ensemble learning approach that used AXoo population and bioclimate as predictors. Tree-based (random forest, extreme gradient boosting) and neural network (single-task deep, multitask deep, mixture of experts) methods were stacked using Firth logistic regression as meta-learner to produce predicted probabilities of the seven Xa genes per Xoo isolate. Gene effectiveness is then measured as the mean predicted probability per gene. For uncertainty measurement of prediction, 95% confidence interval is shown as error bars.
The list of varieties were prepared by curating varieties with whole genome sequence and md-density marker genotyping results using key traits of interest, in this case,the presence or absence of favorable alleles associated with resistance genes for bacterial blight.
To know more about this table, please refer to the QTL Profiles page.