Modelling of fatty acids signatures predicts macroalgal carbon in marine sediments

Abstract

Differentiating between carbon contributors in marine environments is crucial to gaining a deeper understanding of marine carbon sequestration, and some efforts have been made through the application of various approaches. This study proposed a new approach through the use of fatty acid (FA) profiles of six marine macrophytes within three macroalgal lineages, and three coastal angiosperms (mangrove, saltmarsh, and seagrass). We compiled FA profiles (consisting of 84 individuals and 9 classes/groups of FAs) of 544 Australian coastal macrophyte species identified in published reports. The data were gradually screened into three different datasets (full-84FA, reduced-57FA, and reduced-48FA) for analysis to minimise the effects of imbalanced distributions of data on analysis. XGBoost (eXtreme Gradient Boosting) multiclass classification modelling with hyperparameter tuning was applied to reveal the specific FA signatures of each macrophyte lineage. The XGBoost models run across the three datasets generated high model-performance metrics including precision, recall, F-score, and multiclass-AUC, indicating similar performance between the three models with predictive accuracies of 94%, 85%, and 95%, respectively. At the class level, the three models also demonstrated high performance with precision, recall, and F-score values for each lineage above 0.95, except for Rhodophyta, which ranged from around 0.80 to 0.89. Overall, our findings suggest that the XGBoost classifier can reveal the lineage-specific patterns of FAs (carbon-based molecules) that can be implemented to predict and potentially quantify the carbon contributors to marine sediments, and more specifically, to discern macroalgal carbon contributions from those of other coastal macrophytes.

Keywords

• Biomarkers
• XGBoost classifier
• Seaweed
• Carbon sources
• Blue carbon