A classification model for G-to-A hypermutation in hepatitis B virus ultra-deep pyrosequencing reads

Motivation: G -> A hypermutation is an innate antiviral defense mechanism, mediated by host enzymes, which leads to the mutational impairment of viruses. Sensitive and specific identification of host-mediated G -> A hypermutation is a novel sequence analysis challenge, particularly for viral deep sequencing studies. For example, two of the most common hepatitis B virus (HBV) reverse transcriptase (RT) drug-resistance mutations, A181T and M204I, arise from G -> A changes and are routinely detected as low-abundance variants in nearly all HBV deep sequencing samples. Results: We developed a classification model using measures of G -> A excess and predicted indicators of lethal mutation and applied this model to 325 920 unique deep sequencing reads from plasma virus samples from 45 drug treatment-naive HBV-infected individuals. The 2.9% of sequence reads that were classified as hypermutated by our model included most of the reads with A181T and/or M204I, indicating the usefulness of this model for distinguishing viral adaptive changes from host-mediated viral editing.