Validation of the iSNV identification method.

To validate the method described above for identifying iSNVs, metagenomic sequencing was performed from RNA from a plasma sample from a patient with Feibig stage IV infection in a different cohort. This sample had previously undergone SGA and 454 sequencing of the 5′ half of the genome, generating a total of 19 sequences, among which 49 iSNVs were identified (2). Metagenomic sequencing was performed as described above to a moderate depth (45×) in order to compare iSNV detection between the two methods. For each of the 5 iSNVs identified by SGA but not by metagenomic sequencing, we manually inspected the metagenomic sequencing reads.

To assess whether the differences between the two methods were more than could be expected from sampling variance, we applied Fisher’s exact test to each iSNV, including those that failed the duplicate library and strand bias filters. To better understand the resulting P value distribution, we simulated the comparison between the two methods. SGA and metagenomic sequencing results were both modeled as binomial random variates, based on allele frequency. A set of 49 samples was simulated, with coverage and allele frequencies taken from the observed data (the allele frequency was estimated as the mean of those measured by the two methods); the coverage for the two metagenomic libraries was also taken from the data. Metagenomic sequencing had two filters imposed: both alleles had to be seen in two libraries, and both had to be seen on at least one forward and one reverse read. Forward and reverse strand assignment was modeled as a binomial random variate with no strand bias.