Application and challenges of machine learning in microbial remediation: A review of current status and future directions

Microbial remediation is crucial in environmental pollution control. However, targeted intervention is challenging due to the complex and dynamic interactions between microbial communities and external stressors. Machine learning (ML) can be used to deeply analyze the connections between microbial processes and contaminant removal through data mining. Microbial remediation lies at the intersection of microbiology and environmental science, with its diverse scope offering high flexibility for ML applications. Despite the potential of ML, limited attention has been given to its applications within this specific field, and there is a lack of structured reviews to guide the development of ML frameworks in microbial remediation. This review examines the role and current status of ML in microbial remediation. Application modes are presented and compared with a clear hierarchy, including initial monitoring, strategy formulation, and system design. It provides access to established frameworks and alternative solutions to address relevant challenges. Two primary application modes are identified among the seemingly diverse approaches: mapping-based inference and importance-based identification of key agents. The first mode establishes a mapping between two causally linked datasets to predict various outcomes such as remedial effects and microbial growth. Accordingly, the second mode identifies predictors that significantly contribute to mapping accuracies as key microbes or environmental variables. Emerging issues related to the limited accessibility and interpretability are discussed. Finally, using multi-modal learning for pipeline development and applying knowledge graphs (KGs) and a deep reinforcement learning framework to enhance interpretability are proposed as promising solutions.