Integrating metagenomics analysis and machine learning to identify drivers of antibiotic resistance genes abundance in microplastic-contaminated soil

Microplastics (MPs) in soil ecosystems significantly influence antibiotic resistance genes (ARGs) transmission and abundance. However, a holistic understanding of how MP characteristics interact with climatic and edaphic factors to drive ARGs fluctuations remain unclear. By integrating global metagenomic datasets, we compared divergence of soil microbial communities and antibiotic resistomes across MP types and applied interpretable machine learning (ML) techniques to explore ARG dynamics. Results revealed distinct microbial community and resistome patterns associated with MP types, explaining 31.0-36.2 % of the variation in bacterial structure and ARG profiles. Moreover, specific microbial biomarkers for 7 MP types underscore their significant role in structuring communities. Biodegradable MPs (e.g., polybutylene succinate, polyhydroxyalkanoates) exhibited reduced bacterial diversity but higher ARG abundance risk than conventional MPs. Among ML models, Gradient Boosted Decision Trees exhibited superior predictive performance for ARG abundance, with average R² values of 0.98 for training and 0.93 for testing. Driver importance analysis identified bacterial genera (mean contribution: 69.86 %) as the dominant factor in the abundance of primary ARG subtype, followed by climate (13.89 %), soil properties (9.55 %), and MP characteristics (6.70 %). This study provides novel insights into the environmental drivers of ARG dynamics in MP-contaminated soils, highlighting the importance of incorporating climate scenario projections into future ecological risk management strategies for MPs and ARGs.