机器学习助力柑橘黄龙病早期诊断.pdf

1、Machine Learning Enabled Early Diagnosis of HLB in CitrusDr.Yu Wang,Associate ProfessorE-mail:yu.wangufl.eduUniversity of Florida,Institute of Food&Agricultural Sciences,Citrus Research&Education CenterSeptember.25th,2025Early Detection for Citrus HuangLongBing(HLB)Destructive disease of citrusBacte

2、rium:Candidatus Liberibacter asiaticus(CLas)Psyllid vector:Diaphorina citri(D.citri)Citrus production decreased from 250 million boxes in 2005 to 15 million boxes in 2024.qPCR for bacterial detection at 6 monthsSCIENTIFIC MOTIVATIONEarly Detection for Citrus HuangLongBing(HLB)SCIENTIFIC MOTIVATIONCA

3、LIFONIAFLORIDADestroy infected treesEstablish quarantines Prevent further spreadEvaluate novel treatmentsIdentify tolerant/resistant cultivarsUnderstand tree responseBACKGROUNDSamplepreparationInstrumentalanalysisRaw dataprocessingMetabolite identificationComponent 1Component 2Database searchmzCloud

4、METLINNISTHypothesis(tentative confirmation)Targeted metabolomicsPrimaryinformationControlTestmetabolites 1,000(tentatively identified)Nontargeted metabolomics workflowPlant Materials Plant materialsMidsweet sweet orange trees planted in greenhouseBudwoods source:1)Midsweet budwood was grafted onto

5、US-802 rootstock&grown for oneyear2)Seedlings were inoculated with scions from completely pathogen-free&seriously CaLas-infected sour oranges;3)After Passing qPCR test,budwood was cut for graftingSampling:7 weeks post-exposure,10 to 14 leaves were randomly collected from each of 12 individual health

6、y&infected trees.DATA CHALLENGEUHPLC/MS-based nontargeted metabolomics analysishealthyinfectedAnnotated featuresSelectedModel Fitting and Validation Data pre-processing&database searchPCA visualizes differences between citrus trees of healthy group&HLB-affected groupAcquisitionNo HLB-free trees in t