Classification of pmoA amplicon pyrosequences using BLAST and the lowest common ancestor method in MEGAN

The classification of high-throughput sequencing data of protein-encoding genes is not as well established as for 16S rRNA. The objective of this work was to develop a simple and accurate method of classifying large datasets of pmoA sequences, a common marker for methanotrophic bacteria. A taxonomic system for pmoA was developed based on a phylogenetic analysis of available sequences. The taxonomy incorporates the known diversity of pmoA present in public databases, including both sequences from cultivated and uncultivated organisms. Representative sequences from closely related genes, such as those encoding the bacterial ammonia monooxygenase, were also included in the pmoA taxonomy. In total, 53 low-level taxa (genus-level) are included. Using previously published datasets of high-throughput pmoA amplicon sequence data, we tested two approaches for classifying pmoA: a naïve Bayesian classifier and BLAST. Classification of pmoA sequences based on BLAST analyses was performed using the lowest common ancestor (LCA) algorithm in MEGAN, a software program commonly used for the analysis of metagenomic data. Both the naïve Bayesian and BLAST methods were able to classify pmoA sequences and provided similar classifications; however, the naïve Bayesian classifier was prone to misclassifying contaminant sequences present in the datasets. Another advantage of the BLAST/LCA method was that it provided a user-interpretable output and enabled novelty detection at various levels, from highly divergent pmoA sequences to genus-level novelty.