Uncovering dormancy stage predictors in sweet cherry through DNA methylation and machine learning integration

Area of Science:

• Plant Physiology
• Epigenetics
• Computational Biology

Background:

• Sweet cherry (Prunus avium L.) dormancy is crucial for winter survival and spring growth.
• Traditional dormancy prediction methods lack resolution and robustness.
• Epigenetic mechanisms, especially DNA methylation, are key regulators of dormancy transitions.

Purpose of the Study:

• To integrate whole-genome bisulfite sequencing and machine learning (ML) for predicting dormancy stages in sweet cherry.
• To identify DNA methylation markers associated with specific dormancy stages.
• To explore the potential of ML for developing predictive models in perennial species.

Main Methods:

• DNA methylation data from three experiments were classified using Random Forest (RF) and eXtreme Gradient Boosting (XGBoost).
• SHapley Additive exPlanations (SHAP) were used for model interpretability.
• Feature importance was evaluated using Integrated Model consensus across RF, XGBoost, and SHAP metrics.

Main Results:

• Feature selection significantly improved classification accuracy, reaching up to 97.1% in a two-stage model (endodormancy, ecodormancy) using RF.
• SHAP analysis confirmed the ability of selected features to discriminate dormancy stages and identified biologically significant epigenetic markers.
• Key methylation features often co-localized with transposable elements (LTR retrotransposons) and quantitative trait loci (QTLs) for important phenological traits.

Conclusions:

• Combining high-resolution methylation data with interpretable ML provides robust dormancy biomarkers.
• Epigenetic regulation of dormancy may involve chromatin remodeling mediated by transposable elements.
• Findings support the development of non-destructive methylation-based tools for improved phenological management in perennial fruit crops.