Decoding Microbial Strategies: Artificial Intelligence Solutions to Detect Treatment Phenomena and Understand Microbial Metabolism

Abstract

Optimizing microbial metabolism and improving wastewater treatment efficiency require a deep understanding of carbon utilization and advanced microbial monitoring tools. While mixotrophic bacteria offer promising applications in bioplastic production, challenges persist in understanding their metabolic preferences under varying substrate conditions. Similarly, the ability to accurately detect bacteria is essential for preventing sludge bulking and ensuring operational efficiency. This dissertation addresses these challenges by investigating the carbon preferences of a mixotrophic bacterium for enhanced biopolymer synthesis and developing artificial intelligence (AI)-based image processing tools for microbial detection and segmentation in wastewater treatment. Specifically, the metabolic mechanisms and bacteria substrate preferences were explored under different organic (acetate, butyrate) and gaseous (H₂, CO₂, O₂) concentrations, revealing a predominant reliance on organic carbon during growth and a shift toward gaseous substrates during polyhydroxyalkanoate (PHA) accumulation. Then, a deep learning framework was developed to classify filamentous and floc-forming bacteria in microscopic images to enhance microbial detection in wastewater. A rule-based segmentation algorithm was first introduced to generate reliable ground truth labels, followed by a convolutional neural network that was trained on labeled images. The AI model achieved high classification accuracy and demonstrated robust generalization across different microbial morphologies and background intensities. Further optimization was conducted to investigate the effect of training sample size, image resolution, and training epochs, revealing key trade-offs between computational efficiency and segmentation performance. Finally, the effect of utilizing low-resolution annotations to predict high-resolution segmentation was explored to achieve more efficient implementation. By investigating metabolic diversity and AI-driven analysis, this research advances both bioprocess engineering and wastewater treatment and offered scalable methodologies for real-time microbial monitoring, process optimization, and predictive maintenance. The findings contribute to improving biopolymer production strategies and developing automated tools for wastewater treatment facilities targeting more sustainable and efficient biotechnological applications.