Evaluation of genomic prediction accuracy for yield-related traits in Pyropia haitanensis

Pyropia haitanensis, a representative species of intertidal red algae, has significant economic value. However, its traditional breeding cycle is lengthy and inefficient, leading to a shortage of new varieties of P. haitanensis. Genomic selection (GS), an efficient molecular-breeding strategy, holds promise for accelerating the genetic improvement of yield-related traits. To date, no systematic GS study has been reported for red algae. In this study, we used a doubled haploid (DH) population to perform genomic prediction for five yield‑related traits (length, width, fresh weight, days to maturity, and thickness) based on whole-genome sequencing data. We systematically compared the effects of eight statistical models on prediction accuracy and evaluated the effect of different marker densities, ranging from 100 SNPs to the full marker set, under two SNP selection strategies: genome-wide association study (GWAS)-based selection and random selection. The results showed that the heritability of five yield traits ranged from 0.12 (length) to 0.57 (thickness). The random forest (RF) model consistently achieved higher prediction accuracy than the other models across all traits, with accuracies ranging from 0.448 for length to 0.727 for thickness. For all models, prediction accuracy reached a plateau when the number of markers was in the range of 1,000-5,000, with limited gains from further increases in marker density. GWAS-based selection performed better than random selection for most traits at low marker densities (100-1,000 SNPs), while random selection showed higher accuracy for width. In summary, the application of genomic selection of P. haitanensis breeding provides an effective strategy for identifying superior candidate individuals. In practical applications, the optimal model should be selected flexibly according to the genetic architecture of the traits and the available SNP density.

The online version contains supplementary material available at https://doi.org/10.1007/s11032-026-01732-y.