Mapping QTLs for PHS resistance and development of a deep learning model to measure PHS rate in japonica rice

Rice (Oryza sativa L.) is a staple food for more than half of the global population. Preharvest sprouting (PHS), which reduces yield and grain quality, presents a major challenge for rice production. The development of PHS-resistant varieties is a major goal in japonica rice breeding. A deep learning model to automate PHS rate measurement was developed using the YOLOv8 algorithm. The model had high mean average precision (0.974). PHS rate measurements made using the model correlated strongly with manual measurements (R²  =  0.9567). A population of 182 F₈ recombinant inbred lines (RILs) was derived from a cross between the japonica rice cultivars, Junam and Nampyeong. The RIL genotypes at 763 single nucleotide polymorphism markers were determined using a rice target capture sequencing system and used to create a genetic map. The RILs were cultivated in the field (summer season) and the greenhouse (winter season) and their PHS rates were measured in both environments. Quantitative trait loci (QTLs) associated with PHS were present on chromosomes 3, 6, and 7 in the field, and on chromosomes 1, 2, 3, 6, 7, 8, and 11 in the greenhouse. Three QTLs on chromosomes 3, 6, and 7 showed stable effects in both environments. A search for candidate genes in the QTL qPHS6 identified Os06g0317200. This gene encodes a glycine-rich protein resembling qLTG3-1, which controls PHS. The QTLs identified in this study and the deep learning model developed for measuring PHS rates will accelerate the development of rice varieties with enhanced resistance to PHS.