Evaluating performance of metagenomic characterization algorithms using in silico datasets generated with FASTQSim

Anna Shcherbina; Darrell O. Ricke; Nelson Chiu

MetaPhyler does not report results more specific than the Class taxonomy level for the Human 221 dataset, in line with the conservative approach of this algorithm. MetaPhlAn crashes with a 222 segmentation fault on the Human dataset, which most likely is an artifact of the non--host--filtering 223 approach used by this algorithm. 224 225 The algorithms were evaluated based on their ability to correctly map reads and predict relative 226 abundance of the organisms in the data (Figures 2,3) . For the bacterial datasets, Kraken and MetaScope 227 . CC-BY 4.0 International license is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It . https://doi.org/10.1101/046532 doi: bioRxiv preprint classified the highest number of reads correctly for both the genus and species level, and cluster closest 228 to the truth in the dendrogram. However, for the viral datasets, LMAT performed best, classifying the 229 most reads correctly. 230