Machine learning-based metabolomics analysis reveals the early biomarkers for Diplodia stem-end rot in grapefruit caused by Lasiodiplodia theobromae

Abstract

Lasiodiplodia theobromae is a key postharvest pathogen causing Diplodia stem-end rot (SER) disease in grapefruit. While the disease remains quiescent before harvest, its symptoms become evident during the postharvest period. This dormant behavior poses a challenge in managing fruits after harvest. To effectively detect asymptomatic fruit and detect SER disease at an early stage, it's crucial to identify early-stage biomarkers that can serve as disease indicators. In this study, a machine learning-based metabolomics analysis was utilized to identify characteristic metabolites and to elucidate the underlying biosynthetic mechanisms. Six machine learning algorithms were used, and Gradient boosting (GBT) exhibited the highest accuracy identifying early biomarkers such as shikimate, succinic acid, quinic acid, coumaric acid, tyrosine, phenylalanine, and tryptophan. Moreover, dynamic time warping (DTW), was used to investigate the trend of metabolites across timepoints. Our result revealed that the metabolic analysis enabled differentiating infected from non-infected fruits within 1 day, even though symptoms appeared after 7 days of inoculation. Pathway enrichment analysis indicated that three pathways (biosynthesis of plant hormones, phenylpropanoid biosynthesis, and glutamate metabolism) were strongly involved in the defense mechanism. Metabolite mapping analysis showed the behavior of each compound against the pathogen. With these novel strategies, our findings suggest a prediction model for identifying asymptomatic grapefruit affected by Diplodia SER.