Simulation models of microbial community, pH, and volatile fatty acids of anaerobic digestion developed by machine learning

Abstract

Anaerobic digestion (AD) is a promising method for treating high-content organic waste via the microbiome in an anaerobic environment. However, because AD is a complex process involving microbial community (MC), it is essential to determine the relationship between the MC, pH, and volatile fatty acids (VFAs). This study employed five machine learning models: artificial neural network, convolutional neural network, extra trees, extreme gradient boost, and support vector machine to estimate the relationship between MC, pH, and VFAs. Methanoregulaceae, Methanomicrobiaceae, Spirochaetes, pH, acetic acid, isobutyric acid, and isovaleric acid were estimated with high accuracy, with R2 >0.800. The variable importance of the models with the highest accuracy was calculated based on the shapley additive explanations analysis. As a result, the MC estimation models suggested syntrophic acetate oxidation by Spirochaetes, and VFAs-pH estimation models suggested that the high organic loading rate during the start-up phase can lead to isovaleric acid accumulation. In an in-silico test using the machine learning simulator, the change of MC resulted in the critical effects on the acetate accumulation from 500 to 1500 mg/L. The optimal relative abundance of Mehtanoregulaceae and Methanomicrobiaceae for the stable operation was suggested as 15–37 % and 3–15 %, respectively, resulting in the acetic acid concentration below 1000 mg/L.