Machine learning approaches for predicting antibiotic resistance genes abundance changes during biological nitrogen...

Area of Science:

• Environmental microbiology
• Wastewater engineering
• Computational biology

Background:

• Wastewater treatment plants (WWTPs) are reservoirs for antimicrobial agents (AAs), increasing the risk of antibiotic resistance genes (ARGs) transmission.
• The complex interplay of factors and microbial communities makes tracking ARG fate during biological nitrogen removal (BNR) challenging.

Purpose of the Study:

• To apply machine learning (ML) models for predicting ARG and mobile genetic element (MGE) abundance changes during BNR processes.
• To identify critical factors influencing ARG and MGE dynamics.
• To develop data-driven strategies for mitigating ARG transmission risks.

Main Methods:

• Four machine learning models, including Categorical Boosting (CatBoost) and Random Forest (RF), were employed to predict ARG and MGE abundance.
• Feature importance analysis was conducted to identify key drivers of ARG and MGE changes.
• Partial Dependence Plots (PDPs) were used to analyze the relationships between factors and abundance changes.

Main Results:

• The CatBoost model achieved high accuracy (R² = 0.843) in predicting ARG abundance changes, while the RF model excelled in predicting MGE abundance changes (R² = 0.708).
• Mobile genetic element (MGE) abundance was the most significant predictor of ARG abundance changes.
• Microbial community composition (Bacteroidetes, Proteobacteria) and environmental factors (exposure time, pollutant concentration) were critical influencers.

Conclusions:

• Machine learning provides effective approaches for evaluating and controlling ARG transmission risks in WWTPs.
• Understanding the influence of MGEs, microbial communities, and environmental factors is crucial for ARG mitigation.
• A user-friendly interface was developed to guide the optimization of ARG control strategies during BNR processes.