Machine Learning Predicts the Presence of 2,4,6-Trinitrotoluene in Sediments of a Baltic Sea Munitions Dumpsite Using Microbial Community Compositions

• DOI: 10.3389/fmicb.2021.626048
• Publication Date: 2021-09-29T16:16:50Z
• Authors: René Janßen; Aaron J. Beck; Johannes Werner; Olaf Dellwig; Johannes Alneberg; Bernd Kreikemeyer; Edmund Maser; Claus Böttcher; Eric P. Achterberg; Anders F. Andersson; Matthias Labrenz
• Keywords: Trinitrotoluene; Baltic sea

Bacteria are ubiquitous and live in complex microbial communities. Due to differences physiological properties niche preferences among community members, communities respond specific ways environmental drivers, potentially resulting distinct fingerprints for a given state. As proof of the principle, our goal was assess opportunities limitations machine learning detect indicating presence munition compound 2,4,6-trinitrotoluene (TNT) southwestern Baltic Sea sediments. Over 40 variables including grain size distribution, elemental composition, concentration compounds (mostly at pmol⋅g –1 levels) from 150 sediments collected near-to-shore dumpsite Kolberger Heide by German city Kiel were combined with 16S rRNA gene amplicon sequencing libraries. Prediction achieved using Random Forests (RFs); robustness predictions validated Artificial Neural Networks (ANN). To facilitate microbiome data we developed R package phyloseq2ML. Using most classification-relevant 25 bacterial genera exclusively, representing TNT-indicative fingerprint, TNT predicted correctly up 81.5% balanced accuracy. False positive classifications indicated that this approach also has potential identify samples where original contamination no longer detectable. The fact not main drivers composition demonstrates sensitivity approach. Moreover, resulted poorer prediction rates than fingerprints. Our results suggest can predict even minor influencing factors environments, demonstrating discovery events over an integrated period time. Proven environment future studies should ability monitoring general.