Application of Machine Learning to Identify Key Parameters in Chemotactic Bacterial Distributions Near Oil Ganglia in Porous Media

Nonaqueous phase liquids (NAPLs) are pollutants that can persist underground for decades due to their low water solubility. Bioremediation may provide a cheap and effective solution for removing NAPL contaminants underground. Chemotaxis, a mechanism bacteria use to detect and move towards higher chemical concentrations, including NAPLs, may enhance the overall rate of biodegradation. Since bacteria consume NAPLs as food, greater accumulation of chemotactic bacteria near the NAPLs suggests higher efficacy of bioremediation. Our study applies a Machine Learning algorithm to evaluate factors that may influence bacterial transport and chemotaxis in a dual-permeability microfluidic device in the presence of trapped NAPL ganglia. From our experimental observations, we hypothesize that bacterial accumulation is impacted by the dimensions of the micro-pocket, flow rate, time elapsed since the release of bacteria, and locations of NAPL ganglia. Machine Learning analysis suggest the length of the micro-pocket is an important feature for regulating bacterial accumulation. Longer micro-pockets make NAPL gradients more robust because fluid velocity slows down near the NAPL ganglia. Therefore, bacteria may accumulate at higher rates upon sensing these gradients. Linear Regression is a practical starting point to identify trends in numerical variables. In future studies, we plan to test more advanced Machine Learning algorithms, such as neural networks. At the same time, we will obtain more experimental data from microfluidic devices to improve our models. Machine Learning is expected to be a useful tool to understand the transport of chemotactic bacteria in oil-contaminated porous media and improve applications of chemotaxis in bioremediation.