Integrative Machine Learning Approaches for Identifying Loci Associated with Anthracnose Resistance in Strawberry

Anthracnose, predominantly caused by Colletotrichum fructicola, severely reduces yield in Fragaria × ananassa production. We assessed ensemble machine learning (ML) frameworks to improve genomic prediction (GP) of resistance using a training population of 300 individuals from six full-sib families. Genotyping with the Axiom® 50K FanaSNP array and phenotyping by AUDPC after artificial inoculation enabled evaluation of five algorithms—G-BLUP, LASSO, LightGBM, Random Forest, and XGBoost—combined with informed feature selection and resampling-based data augmentation (3×, 5×). Ensemble ML models consistently outperformed linear approaches, with Random Forest, LightGBM, and XGBoost achieving the highest accuracies. Marker prioritization revealed that a reduced SNP panel of ~200 markers provided near-maximal predictive performance (R2 up to 0.991), demonstrating that compact subsets can support cost-efficient GP. Data augmentation, implemented through the resampling of existing observations rather than the creation of new alleles, improved statistical power and model stability under limited sample sizes. Collectively, this study demonstrates that (i) ensemble ML models deliver superior accuracy for predicting polygenic resistance, (ii) small SNP panels can achieve high efficiency, and (iii) augmentation enhances robustness in resource-constrained breeding populations. These findings establish a scalable and breeder-oriented GP pipeline to accelerate the development of anthracnose-resistant strawberry cultivars.