Classifying Asian Rice Cultivars (Oryza sativa L.) into Indica and Japonica Using Logistic Regression Model with Publicly Available Phenotypic Data

This article introduces how to implement the logistic regression model (LRM) with phenotypic variables for classifying Asian rice (Oryza sativa L.) cultivars into two pivotal subpopulations, indica and japonica. This study took advantage of publicly available data attached to a previous paper. The classification accuracy was assessed using an area under curve (AUC) of a receiver operating characteristic (ROC) curve. Given 24 phenotypic variables for 280 indica/japonica accessions, the LRMs were fitted with up to six phenotypic variables of all possible combinations; the highest AUC accounts for 0.9977, obtained with six variables including panicle number per plant, seed number per panicle, florets per panicle, panicle fertility, straighthead susceptibility and blast resistance. Overall, the more variables there are, the higher the resulting AUCs are. The ultimate purpose of this study is to demonstrate the indica/japonica prediction ability of the LRM when applied to unclassified Asian rice cultivars. To estimate the indica/japonica prediction accuracy, ten-fold cross-validations were conducted 100 times with the 280 indica/japonica accessions using the LRM with parameters that yielded the highest AUC. The resulting prediction accuracy accounted for 0.9779. This suggests that the LRM promises to be a highly effective indica/japonica prediction tool using phenotypic variables in Asian cultivated rice.