Machine Learning-Assisted Discrimination of Organ-Specific Radish Pigmentation Through Targeted Phytochemical Profiling

Phytochemical differentiation between radish leaves and roots has implications for crop utilization and quality-oriented breeding, yet its association with tissue pigmentation and genotype remains unresolved. We quantified 18 glucosinolate (GSL) compounds, total phenolic content (TPC), and antioxidant capacity, including ABTS and DPPH radical scavenging activities, in the leaves and roots of 23 early-maturing radish genotypes. Chemometrics, machine learning (ML), and Shapley additive explanations (SHAP) resolved pigmentation-associated phytochemical signatures. Total GSL concentrations in roots were 3.1 times higher than in leaves, while leaves showed 3.3 times higher TPC and greater antioxidant activity. Dark green leaves and white roots were GSL-enriched, while light green leaves and red or mixed-color roots exhibited higher TPC and ABTS activity. Within-group variation distinguished genotypes with high GSL or strong phenolic and antioxidant profiles. Under genotype-grouped cross-validation, elastic-net logistic regression classified leaf and root pigmentation with balanced accuracies of 0.97 and 0.95. SHAP prioritized gluconasturtiin, glucoerucin, and sinalbin in leaves, and TPC and glucoraphasatin as the leading contributors in roots. Collectively, these findings provide trait-specific germplasm resources and candidate phytochemical discriminators for radish quality improvement.