Structural Analysis of Tobacco Rhizosphere Soil Microbial Communities Based on Metagenomics and Deep Learning for Association with Disease Resistance

Abstract

The rhizosphere microbiome, often termed the plant’s “second genome”, plays a pivotal role in regulating plant health and disease resistance. This study integrated metagenomic sequencing and deep learning to systematically compare the composition, diversity, and functional metabolism of microbial communities in healthy (NB) and diseased (NF) tobacco rhizosphere soils. Using BGISEQ-500 sequencing, 18 soil samples were analyzed, yielding 8.63 million clean reads and 6.27 million non-redundant genes. Taxonomic profiling revealed Proteobacteria (35%), Actinobacteria (16%), Firmicutes (10%), and Bacteroidetes (7%) as dominant phyla. Significant structural disparities in α-diversity indices (e.g., Shannon and Simpson) and β -diversity (PCoA) were observed between NB and NF groups (ANOVA, p < 0.05). LEfSe analysis identified 181 and 240 biomarkers in NB and NF, respectively, with healthy soils enriched in Gemmatimonadetes and Sphingomonadaceae, while diseased soils were dominated by Deltaproteobacteria and Rubrobacteraceae. Functional annotation highlighted the enrichment of sulfur metabolism (ko00920), terpenoid biosynthesis (ko00900), and antibiotic synthesis (ko01130) pathways in NB, whereas NF exhibited upregulated lipopolysaccharide biosynthesis (ko00540) and flagellar assembly (ko02040). A deep learning model (H2O framework, two hidden layers) achieved perfect classification (AUC = 1.0) and identified 10 microbial biomarkers as robust indicators of soil health. These findings elucidate the linkage between rhizosphere microbiome dynamics and tobacco disease resistance, providing a methodological framework for ecological disease control and precision agriculture. The study highlights the potential of microbial biomarkers in guiding sustainable agricultural practices and reducing reliance on chemical pesticides.