Machine learning to predict dynamic changes of pathogenic Vibrio spp. abundance on microplastics in marine...

Area of Science:

• Environmental microbiology
• Marine ecology
• Computational toxicology

Background:

• Microplastics are ubiquitous in marine ecosystems.
• Pathogenic microorganisms, including Vibrio spp., can colonize microplastics, potentially acting as vectors for disease transmission.
• Vibrio spp. pose significant risks for human health through foodborne illnesses.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the relative abundance of Vibrio spp. on microplastics.
• To identify key environmental factors influencing Vibrio spp. presence on microplastics.
• To provide a basis for monitoring systems assessing microplastic-related health risks.

Main Methods:

• Utilized datasets from Chinese estuary and mariculture zones.
• Developed and compared five machine learning models, including deep neural network (DNN) and RandomForest.
• Employed SHapley Additive exPlanations (SHAP) for factor identification and feature selection.

Main Results:

• DNN and RandomForest models demonstrated superior predictive performance for Vibrio spp. abundance.
• Salinity and temperature were identified as the primary environmental drivers.
• Model performance was robust across different data sources and processing methods.
• Feature selection enhanced prediction accuracy.

Conclusions:

• Machine learning models, particularly DNN and RandomForest, are effective tools for predicting Vibrio spp. on microplastics.
• Environmental factors like salinity and temperature are crucial for understanding pathogen dynamics on microplastics.
• The findings support the development of monitoring systems for microplastic-associated health risks in marine environments and for human populations.