Phylogeny-Aware Deep 1-Dimensional Convolutional Neural Network for the Classification of Metagenomes

This paper evaluates a novel approach to the integration of biological domain knowledge relating to the natural evolutionary structure of microbial community data to classifying 16S rDNA sequence samples. Specifically, we evaluate the use of phylogenetic trees in addition to amplicon sequence variant abundance in samples for the classification of a processed cattle metagenomics data set using machine learning. Further to this, we employ a class activation map of the network when applied to specific exemplars to determine, firstly, the relevance of higher level taxonomic data, and secondly, the most relevant taxa in determining the classification, according to the classifier.