Feasibility study of machine learning to explore relationships between antimicrobial resistance and microbial community structure in global wastewater treatment plant sludges

Yi Li; Cuicui Tao; Shuyin Li; Wenxuan Chen; Dafang Fu; Chad T Jafvert; Tengyi Zhu

doi:10.1016/j.biortech.2024.131878

Feasibility study of machine learning to explore relationships between antimicrobial resistance and microbial community structure in global wastewater treatment plant sludges

Bioresour Technol. 2024 Nov 25:131878. doi: 10.1016/j.biortech.2024.131878. Online ahead of print.

Authors

Yi Li¹, Cuicui Tao¹, Shuyin Li¹, Wenxuan Chen², Dafang Fu³, Chad T Jafvert⁴, Tengyi Zhu⁵

Affiliations

¹ School of Environmental Science and Engineering, Yangzhou University, Yangzhou 225127, Jiangsu, China.
² Department of Applied Microbial Ecology, Helmholtz Centre for Environmental Research (UFZ), 04318 Leipzig, Germany; School of Civil Engineering, Southeast University, Nanjing 210096, China.
³ School of Civil Engineering, Southeast University, Nanjing 210096, China.
⁴ Lyles School of Civil Engineering, and Environmental & Ecological Engineering, Purdue University, West Lafayette, IN 47907, USA. Electronic address: jafvert@ecn.purdue.edu.
⁵ School of Environmental Science and Engineering, Yangzhou University, Yangzhou 225127, Jiangsu, China. Electronic address: tyzhu@yzu.edu.cn.

PMID: 39603473
DOI: 10.1016/j.biortech.2024.131878

Abstract

Wastewater sludges (WSs) are major reservoirs and emission sources of antibiotic resistance genes (ARGs) in cities. Identifying antimicrobial resistance (AMR) host bacteria in WSs is crucial for understanding AMR formation and mitigating biological and ecological risks. Here 24 sludge data from wastewater treatment plants in Jiangsu Province, China, and 1559 sludge data from genetic databases were analyzed to explore the relationship between 7 AMRs and bacterial distribution. The results of the Procrustes and Spearman correlation analysis were unsatisfactory, with p-value exceeding the threshold of 0.05 and no strong correlation (r > 0.8). In contrast, explainable machine learning (EML) using SHapley Additive exPlanation (SHAP) revealed Pseudomonadota as a major contributor (39.3 %∼74.2 %) to sludge AMR. Overall, the application of ML is promising in analyzing AMR-bacteria relationships. Given the different applicable occasions and advantages of various analysis methods, using ML as one of the correlation analysis tools is strongly recommended.

Keywords: Antibiotic resistance genes; Explainable machine learning; Metagenomics; Microbial community; Wastewater sludge.