Developing Physiologically Compatible Electron Donors for Reductive Dechlorination by Dissimilatory Iron-Reducing Bacteria Using Machine Learning

Abstract

Targeted biological stimulation of carbon sources presents considerable potential for enhancing dehalogenation efficiency at sites contaminated with halogenated hydrocarbons. Combining a natural cellulose-rich carbon source with iron and humic acid has been shown to accelerate reductive dechlorination by dissimilatory iron-reducing bacteria (DIRB) by increasing electron flow pathways. However, organic carbon release in natural environments involves complex interactions among carbon source types, electron transfer, and microbial metabolic activities, making traditional methods insufficient for optimizing carbon sources to accelerate microbial reductive dehalogenation. This study applies machine learning (ML) approaches to elucidate the biocompatibility between carbon source materials and the functional DIRB (Shewanella oneidensis MR-1). Biostimulation conditions and biostimulatory genomic data were used as input variables, with dechlorination effect as the output. The gradient boosting decision tree (XGB) outperformed the random forest (RF), artificial neural network (ANN), and support vector machine (SVM) in assessing the biological dechlorination potential. Feature importance analysis using the optimized XGB model highlighted carbohydrate metabolism and energy metabolism as the primary factors influencing the dechlorination of S. oneidensis MR-1. Insights from ML guided the development of a custom carbon source with higher acetic acid content, leading to a 22% improvement in dechlorination rate and a ∼60–82% reduction in costs. This approach provides a robust framework for designing compatible carbon sources for contaminated sites, grounded in an understanding of microbial physiological functions.