Abstract
The development of modern and affordable sequencing technologies has allowed the study of viral populations to an unprecedented depth. This is of particular interest for the study of within-host RNA viral populations, where variation due to error-prone polymerases can lead to immune escape, antiviral resistance and adaptation to new host species. Methods to sequence RNA virus genomes include reverse transcription (RT) and polymerase chain reaction (PCR). RT-PCR is a molecular biology technique widely used to amplify DNA from an RNA template. The method itself relies on the in vitro synthesis of copy DNA from RNA followed by multiple cycles of DNA amplification. However, this method introduces artefactual errors that can act as confounding factors when the sequence data are analysed. Although there are a growing number of published studies exploring the intra- and inter-host evolutionary dynamics of RNA viruses, the complexity of the methods used to generate sequences makes it difficult to produce probabilistic statements about the likely sources of observed sequence variants. This complexity is further compounded as both the depth of sequencing and the length of the genome segment of interest increase. Here we develop a bayesian method to characterise and differentiate between likely structures for the background viral population. This approach can then be used to identify nucleotide sites that show evidence of change in the within-host viral population structure, either over time or relative to a reference sequence (e.g. an inoculum or another source of infection), or both, without having to build complex evolutionary models. Identification of these sites can help to inform the design of more focussed experiments using molecular biology tools, such as site-directed mutagenesis, to assess the function of specific amino acids. We illustrate the method by applying to datasets from experimental transmission of equine influenza, and a pre-clinical vaccine trial for HIV-1.