Machine Learning Prediction of Nitrification From Ammonia- and Nitrite-Oxidizer Community Structure

• Published in: Frontiers in microbiology 2022-07, Vol.13, p.899565-899565
• Main Authors: Lee, Conard, Amini, Fatemeh, Hu, Guiping, Halverson, Larry J.
• Format: Article
• Language: English
• Subjects: ammonia-oxidizers; machine learning; Microbiology; microbiome; nitrification; nitrifiers
• ISSN: 1664-302X; 1664-302X
• DOI: 10.3389/fmicb.2022.899565

Accurately modeling nitrification and understanding the role specific ammonia- or nitrite-oxidizing taxa play in it are of great interest and importance to microbial ecologists. In this study, we applied machine learning to 16S rRNA sequence and nitrification potential data from an experiment examining interactions between cropping systems and rhizosphere on microbial community assembly and nitrogen cycling processes. Given the high dimensionality of microbiome datasets, we only included nitrifers since only a few taxa are capable of ammonia and nitrite oxidation. We compared the performance of linear and nonlinear algorithms with and without qPCR measures of bacterial and archaea ammonia monooxygenase subunit A ( amoA ) gene abundance. Our feature selection process facilitated the identification of taxons that are most predictive of nitrification and to compare habitats. We found that Nitrosomonas and Nitrospirae were more frequently identified as important predictors of nitrification in conventional systems, whereas Thaumarchaeota were more important predictors in diversified systems. Our results suggest that model performance was not substantively improved by incorporating additional time-consuming and expensive qPCR data on amoA gene abundance. We also identified several clades of nitrifiers important for nitrification in different cropping systems, though we were unable to detect system- or rhizosphere-specific patterns in OTU-level biomarkers for nitrification. Finally, our results highlight the inherent risk of combining data from disparate habitats with the goal of increasing sample size to avoid overfitting models. This study represents a step toward developing machine learning approaches for microbiome research to identify nitrifier ecotypes that may be important for distinguishing ecotypes with defining roles in different habitats.