Agricultural researchers have developed an explainable machine-learning framework to predict key agronomic traits, classify subspecies, and rank potential breeding crosses in rice. The system addresses common barriers in genomic selection, where adoption has frequently been limited by the black-box nature of complex predictive models and an absence of tools that translate statistical outputs into actionable crossing recommendations for breeders.
The investigation utilized the publicly available 1,000 Rice Core Accessions SNP panel, designated as 1k-RiCA, which provided genotype data across 965 loci for 353 indica rice accessions. The team designed the framework around three explicit objectives: predicting flowering time and plant height from marker profiles, classifying accessions into subspecies subgroups, and ranking all possible pairwise parent crosses according to predicted progeny performance.
To evaluate regression performance for quantitative traits, the authors compared ridge regression, random forest, and extreme gradient boosting algorithms. XGBoost delivered the highest predictive accuracy for flowering time, recording a test coefficient of determination of 0.69 alongside a root mean square error of 2.52 days when predicting developmental timing directly from the mid-density marker data.
For plant height, the most effective architecture was a random forest model that incorporated flowering time as a predictive covariate combined with univariate feature selection. This optimized configuration achieved a test coefficient of determination of 0.55 and a root mean square error of 5.94 centimeters, demonstrating the value of integrating related physiological covariates into genomic predictions.
Subspecies subgroup classification was evaluated by contrasting random forest and XGBoost classifiers across the accession panel. Random forest achieved the superior classification outcome, securing an overall accuracy of 76.7 percent and a macro-averaged area under the receiver operating characteristic curve of 0.82, establishing reliable separation among the evaluated indica subgroups.
In contrast to vegetative traits, grain yield was deliberately excluded from genomic prediction models after formal evaluation using ridge-regression best linear unbiased prediction revealed an R² of 0.115 ± 0.061. This near-zero predictive signal aligned with substantial genotype-by-environment interaction effects, leading researchers to retain grain yield solely as a descriptive metric within composite crossing scores.
To ensure model transparency, the researchers conducted SHapley Additive exPlanations analysis, successfully identifying the specific genetic markers driving individual model predictions. The framework systematically evaluated and ranked all 62,128 possible pairwise parent crosses among the accessions, generating a prioritized shortlist to guide selective breeding decisions.
The complete pipeline was implemented as an interactive web application supporting custom dataset uploads, automated data preprocessing, explainable visualization through SHAP metrics, and exportable cross rankings. This deployment provides practical, statistically grounded decision support for operational breeding programs using mid-density marker panels without requiring costly whole-genome sequencing.
Subject of Research: Explainable machine learning for genomic prediction and parent cross ranking in rice breeding
Article Title: Explainable machine learning for genomic prediction, subgroup classification, and optimal parent cross ranking in rice breeding using 1k-RiCA SNP data
Article References: Singh, G. (2026). Explainable machine learning for genomic prediction, subgroup classification, and optimal parent cross ranking in rice breeding using 1k-RiCA SNP data. Scientific Reports. https://doi.org/10.1038/s41598-026-79467-1
Image Credits: AI Generated
DOI: 10.1038/s41598-026-79467-1
Keywords: genomic selection, rice breeding, 1k-RiCA, SNP markers, XGBoost, random forest, flowering time, plant height, grain yield, SHAP analysis, subspecies classification, parent cross ranking, indica rice, ridge regression, rrBLUP, genotype by environment interaction, decision support systems
Cite Scienmag News
Alan Morgan. (November 22, 2026). Explainable Machine Learning Framework Accurately Predicts Key Rice Traits and Crosses. Scienmag. https://scienmag.com/explainable-machine-learning-framework-accurately-predicts-key-rice-traits-and-crosses/
Alan Morgan. "Explainable Machine Learning Framework Accurately Predicts Key Rice Traits and Crosses." Scienmag, 22 November 2026, https://scienmag.com/explainable-machine-learning-framework-accurately-predicts-key-rice-traits-and-crosses/. Accessed 22 November 2026.
Alan Morgan. "Explainable Machine Learning Framework Accurately Predicts Key Rice Traits and Crosses." Scienmag. November 22, 2026. https://scienmag.com/explainable-machine-learning-framework-accurately-predicts-key-rice-traits-and-crosses/








