An automated maximum likelihood method for classifying virus sequences

Background: The classification of virus sequences is essential to support epidemiological surveillance and patient care. The "Rega typing framework", an automated classification method that applies Neighbor-Joining (NJ) phylogenetics, has been shown an effective and popular tool to classify various viral pathogens. However, this method has some important limitations: (a) its scoring strategy evaluates the quality of the assignment indirectly, (b) the procedure is non-deterministic and (c) its cubic computational complexity prohibits the use of large reference sets. Methods: An alternative automated procedure for virus classification, based on maximum likelihood (ML) phylogenetic placement (i.e. pplacer), was developed and integrated in the "Rega typing framework". A score, that represents the confidence of the query sequence's location in a particular clade, was composed. The procedure assigns a classification by selecting the clade with the highest score, if that score exceeds a calibrated threshold. Results: The ML method was validated on a large dataset of hepatitis C virus sequences (Los Alamos HCV database, n=20016, >=800 base pairs per sequence) and compared to the NJ method that was applied on the same dataset. This comparison demonstrates a high level of concordance between the results for the ML and NJ method (97.367%). Conclusion: This research demonstrates the potential of phylogenetic placement to classify virus sequences. The method addresses several limitations of the NJ approach: it delivers (a) a score that directly signifies classification confidence, (b) a deterministic classification approach and (c) a linear time complexity with respect to the number of reference sequences.status: accepte