Machine Learning-Based Heading Date QTL Detection in Rice

Quantitative trait locus (QTL) analysis is a powerful approach for identifying variants associated with the phenotypic variation of complex traits. However, selecting optimal methods and pre-processing steps require considerable time and effort. In this study, we demonstrated applicability and replicability of machine learning (ML) models in QTL analysis by evaluating their performance in comparison with conventional QTL analysis methods using 142 recombinant inbred lines derived from two japonica rice cultivars, Koshihikari and Baegilmi. Random forest and gradient boosting models showed the highest predictive accuracy, and consistently identified three QTLs associated with heading date: qDTH3, qDTH6, and qDTH7. Moreover, ML-based QTL analysis detected minor-effect qDTH10, where Koshihikari allele promoted heading date when combined with Koshihikari alleles of qDTH6 and qDTH7. These results demonstrate the applicability of ML models in QTL analysis on bi-parental mapping population in rice.