Cost-effective colour feature-based rapid prediction of secondary metabolites in Dendrobium nobile with insights from MALDI-MSI and transcriptomic analysis.
TL;DR
A novel rapid assessment approach was developed by integrating colour features with a stacked ensemble machine learning model, which demonstrated superior efficiency and cost-effectiveness compared with traditional methods and revealed the potential association mechanism among colours, metabolites, and genes of D. nobile.
Abstract
Background
AND
Purpose
Secondary metabolite content is a key indicator of medicinal plant quality. However, traditional quantitative analysis methods are often time-consuming and involve the use of toxic reagents. This study aims to address the challenge of rapidly and cost-effectively detecting the content of various secondary metabolites in the medicinal parts of Dendrobium nobile, and proposes an innovative solution based on multi-source color feature data and stacked ensemble learning. It also combines spatial metabolomics and transcriptomics for joint analysis to reveal the underlying potential biological mechanisms.
Methods
AND
Results
A novel rapid assessment approach was developed by integrating colour features with a stacked ensemble machine learning model, which demonstrated superior efficiency and cost-effectiveness compared with traditional methods. The model was validated using multi-year stem samples of D. nobile from diverse origins and harvest times, demonstrating strong generalisability and high predictive accuracy (R²: 0.73-0.88). SHapley Additive exPlanations analysis indicated that warm colour features in old stems contributed strongly to flavonoid and phenolic acid predictions, whereas green colour features in young stems contributed to alkaloid prediction. The spatial distributions of 20 flavonoids, 14 phenolic acids, and 10 alkaloids were significantly correlated with key colour features. Spatial metabolomics further confirmed these patterns, showing progressive warm colour shifts and metabolite redistribution during maturation. Furthermore, transcriptome-metabolite correlation analysis further focused on the phenylpropanoid, flavonoid, flavone and flavonol, isoquinoline alkaloid, and shikimate biosynthesis pathways, which suggested that genes involved in these pathways might take part in regulating the synthesis of the three above-mentioned compounds.
Conclusion
This study successfully developed a rapid and cost-effective quality assessment method based on colour features and machine learning, which demonstrated high reliability when applied to D. nobile. These findings not only reveal the potential association mechanism among colours, metabolites, and genes of D. nobile, but also provide a technical framework for the rapid quality assessment of medicinal plants.