Machine learning application reveal dynamic interaction of polyphosphate-accumulating organism in full-scale wastewater treatment plant

Abstract

Enhanced biological phosphorous removal (EBPR) represents one of the most widely used wastewater systems for treating phosphorous-bearing waste streams. This study aimed to advance the current understanding of the diversity and interactions of polyphosphate-accumulating organisms (PAOs), playing a key role for the EBPR performance. Frequently occurring and predominant organisms consisted of PAOs such as Tetrasphaera, Candidatus Accumulibacter, Acinetobacter, Pseudomonas, and glycogen-accumulating organisms (GAOs) such as Defluvicoccus, Microlunatus, Nakamurella, and Kineosphaera. The degree of immigrating PAOs/GAOs carried with the influent was associated with their succession within the EBPR processes. Ca Accumulibacter and Nakamurella were transported to a greater degree with the influents of combined and separate sewers, respectively. Although no significant shifts in the PAO-GAO community structure were observed across seasons using conventional ordination-based analyses, machine learning (ML) modeling using the PAO-GAO microbiome data could successfully predict the EBPR system's environmental condition (water temperature) with linear and non-linear models (R2 of 0.4–0.7). ML-based feature importance analysis identified two specific species populations, Ca Accumulibacter (PAO) and Defluvicoccus (GAO), strongly correlated with high water temperatures, in contrast to the existing understanding based on laboratory experiments. The positive association among Ca Accumulibacter, Defluvicoccus, and the water temperature in full scale EBPR processes was validated with microbial association network analysis as well as the ML-based results. Overall, our ML-enabled analysis could provide new findings about the interplay between PAOs and environmental factors, which have practical implications for sustainable design and operation of full scale EBPR processes.