Machine-Learning-Assisted Auto-Fluorescence Cell-Sorting-Based Fast Differentiation and Quantification of Extracellular Polymeric Substances in the Microalgae and Bacteria Consortium.

Extracellular polymeric substances (EPS) are crucial in microalgae-bacteria consortia (MBC)-based wastewater treatment. Revealing the roles of EPS depends on reliable and accurate differentiation, extraction, and quantification of EPS from microalgae and bacteria in the MBC. Autofluorescence-based cell sorting (AFCS) coupled with thermal extraction can achieve the above goals, but it is time-consuming and costly, and invoves unignorable EPS loss (0.4-3.9%) during cell sorting. In this work, the raw data of microalgae (3.42 ± 0.14 mg/g TSS - 52.38 ± 0.47 mg/g TSS) and bacteria (2.95 ± 0.17 mg/g TSS - 48.94 ± 0.21 mg/g TSS) EPS from a series of MBC are obtained using AFCS with thermal extraction. Machine learning regression models are applied for microalgae and bacteria identification, and EPS concentration simulation and prediction. Results showed that, based on the total EPS concentrations and the fluorescence scattering characteristics (obtained from flow cytometry) of the MBC before and after heating, machine learning regression models enabled accurate prediction of EPS concentrations from microalgae (k-Nearest Neighbors, R2 = 0.975, P < 0.001) and bacteria (eXtreme Gradient Boosting, R2 = 0.953, P < 0.001) in the MBC, without cell sorting or EPS loss. This study provides an innovative method for rapid and accurate quantification of total EPS in microalgae and bacteria without requiring physical separation of the MBC, thereby laying a crucial foundation for in-depth analysis of EPS-based interaction mechanisms and functional regulation of microalgae and bacteria within the MBC during wastewater treatment.