Targeted Profiling and Interpretable Machine Learning Reveal Crop-Use-Associated Glucosinolate Signatures in Brassica juncea

This is the authors' abstract. We don't add key points for this paper.

Background Brassica juncea has morphologically and agronomically diverse cultivated forms used for vegetative, root, and seed production and consumption. Although glucosinolate (GSL) composition varies substantially within B. juncea , the extent to which this crop-use diversification is reflected in leaf GSL composition remains underexplored. This study aimed to characterize leaf GSL variation among accessions representing vegetative (VM), root (RM), and seed mustard (SM) types, and to identify GSL features contributing to crop-use type discrimination. Results The results showed significant differences at both the crop-use forms and accession levels in B. juncea . Aliphatic GSL accounted for more than 98% of the total GSL in all three mustard-types, with sinigrin as the predominant compound. RM accessions exhibited lower total GSL than VM and SM, but comparatively higher 4-hydroxyglucobrassicin and neoglucobrassicin. PCA positioned RM in a distinct region, whereas SM overlapped substantially with the broader VM distribution. Pairwise machine learning (ML) further supported this pattern, with linear support vector machine and random forest models discriminating RM from VM and SM, respectively, while VM and SM exhibited weaker separation. Shapley additive explanations (SHAP) analysis detected recurrent contributions of 4-hydroxyglucobrassicin, neoglucobrassicin, sinigrin, glucocheirolin, and related GSLs to RM-associated differentiation. Conclusion The findings establish a biochemical framework for B. juncea germplasm characterization and support the prioritization of contrasting genetic resources and GSL signatures for future breeding and mustard improvement.