Amplicon sequencing, especially of RNA viruses, requires additional bioinformatics processing to ensure maximum quality of the resulting data.
In RT-PCR, a reverse transcriptase enzyme first generates cDNA molecules using the RNA molecules in the sample as templates, before amplifying the cDNA sequences using a DNA polymerase enzyme during PCR. These amplified cDNA sequences are then further processed to generate the sequencing libraries. Both of these enzymes can potentially introduce an incorrect base into a sequence, generating a position where the resulting sequence does not match the sequence in the sample -- that is, an error.
Reverse transcriptases exhibit error rates that are multiple orders of magnitude higher than those of DNA polymerases.
When large numbers of nucleic acid molecules are present in a reaction, these individual misincorporation errors are largely uncorrelated and appear at very low frequencies, so that they are typically ignored by variant callers.
However, when the number of incoming nucleic acid molecules is small, such as for a low-titer virus sample, an error that occurs during the RT step or early in the PCR reaction can, as a result of sampling noise, be amplified to high frequencies in the resulting sequencing libraries. When the variant caller encounters such a position, it will be treated as a sequence variant, since it is a true sequence variant in the context of the library provided to the instrument. As a result, these artifactual sequence variants often have high allele frequency and very good quality scores, which makes them very difficult to detect and remove. This can result in a false positive variant call that, at a sufficiently high allele frequency, will also be incorporated into the consensus sequence. It is also possible for a true sequence variant to have its allele frequency depressed by this same process (if the error results in a reversion to the reference sequence), but this is much less common.
Since it is difficult to identify enzyme-introduced false variants after the fact, we instead take a pre-emptive approach to ensuring data quality. As noted above, sampling noise as a function of molecular abundance is the mechanism responsible for boosting of the frequency of individual enzymatic errors into artifactual variants, and therefore the magnitude of this effect is largely a function of the concentration of the nucleic acids in the reaction. Therefore, the software first attempts to determine whether there is sufficient sample material present before proceeding with variant calling and consensus sequence generation.
To determine this, the software takes advantage of the fact that the probability of each amplicon being amplified is a function of the nucleic acid concentration, with higher concentrations leading to a higher probability of amplification. By counting the observed proportion of amplicons with detectable sequence coverage, we can estimate this probability and compare it to an experimentally-determined threshold that corresponds to the minimum concentration needed to produce reliable variant calls.
To compute this, we calculate the number of amplicons with at least 1x coverage for at least 90% of the non-overlapping portion of the amplicon sequence. The 1x coverage threshold used here is fixed and independent of the minimum read coverage depth for consensus sequence generation which defaults to 10x. The number of amplicons that meet this threshold is then divided by the total number of amplicons in the experiment, which is the number of amplicons whose location falls in reference sequences selected for short read alignment. If the resulting fraction is at least 80%, the sample is considered to have sufficient material for accurate variant calling and the variant calling and consensus sequence generation steps are performed. If it is below this threshold, the sample is not processed further to avoid spurious variant calls. The user can override the 80% threshold in the "Minimum percentage of amplicons with at least 90% coverage ≥ 1x to enable variant calling and consensus sequence generation" control in the "Advanced Workflow Settings" section. See App Settings.