Towards a 16S rRNA Sequence Analysis Pipeline: Exploring Statistical Methods for Taxonomic Assignment

Using 16S rRNA gene sequencing data is a fast, inexpensive method of performing taxonomic classification on prokaryotes. Many statistical methods have been developed to do such classifications. However, no existing tools for 16S analysis are geared towards long-read data. Long-read data is becoming increasingly accessible, and makes getting full-length 16S gene data significantly more feasible. We explore statistical methods in taxonomic assignment towards the development of a 16S analysis pipeline focused on long-read, full-gene data. We focus on the RDP Classifier, Bayesian Lowest Common Ancestor (BLCA), and Hidden Markov Model-based Utra-Fast OTU tools and assess their effectiveness on a testing set of Proteobacteria. We find that, when using continuous assignment, BLCA performs very similarly to RDP Classifier when using either bootstrap confidence scores or posterior probabilities to perform assignment. We conclude that BLCA’s novel Bayesian method shows great promise for growth and potential inclusion in a long-read, full-gene 16S analysis pipeline.