Machine learning classification of microbial community compositions to predict anthropogenic pollutants in the Baltic Sea

10.18453/rosdok_id00002897 Janßen, René René Janßen http://d-nb.info/gnd/1225131375 https://orcid.org/0000-0002-0605-6860 Machine learning classification of microbial community compositions to predict anthropogenic pollutants in the Baltic Sea Universität Rostock 2020 004 Data processing Computer sciences 570 Life science 2020 en http://purl.uni-rostock.de/rosdok/id00002897 urn:nbn:de:gbv:28-rosdok_id00002897-8 Microbial communities react rapidly and specifically to changing environments, indicating distinct microbial fingerprints for a given environmental state. Machine learning with community data predicted the Baltic Sea-detected pollutants glyphosate and 2,4,6-trinitrotoluene, using the developed R package “phyloseq2ML”. Predictions by Random Forest and Artificial Neural Network were accurate. Relevant taxa were identified. The interpretability of machine learning models was found of particular importance. Microbial communities predicted even minor influencing factors in complex environments.