This is a demonstration environment. It may contain non-accurate test data and should not be used for real-world applications. Data will be deleted regularly.

Processing

After upload, every sample goes through the following steps. Their results appear on the sample’s page; generated files are published next to the submitted ones.

  1. Metadata checks: required fields are present, dates are in the format YYYY-MM-DD and not in the future, the collection end date is not before the start date, numbers are numbers, and picklist values such as the country are among the accepted values.
  2. File checks: the sample has one BAM file, or one VCF file with one depth file. The files must be readable and use the Wuhan-Hu-1 reference; the depth file must list every position of the reference.
  3. Dehosting check (BAM files): Deacon compares the reads with the human genome, using the same index and parameters as Loculus’ raw-reads processing. If more than 10 % of the reads or more than 300 kbp match, the submitter gets a warning stating the measured values. Such a sample should not be approved: remove the host reads and upload the sample again.
  4. SNV frequencies and depth (BAM files): freyja variants (samtools mpileup and iVar) derives the SNV frequencies, published as a VCF file, and the depth per site, published as a TSV file.
  5. Coverage: the mean depth and the share of the genome covered at depth 10 or more are computed from the depth per site.
  6. Lineage prevalence: Freyja estimates the abundance of Pango lineages from the SNV frequencies and the depth per site. It uses lineage barcodes from a fixed Freyja version; the sample’s page states the Freyja version and the date of the barcodes. Lineages with an abundance of at least 1 % are shown (at most 10, the rest summed up as “Other”) as a text summary and a bar chart; all estimated abundances are published as a TSV file and in the metadata field lineage_prevalence_full. The lineage with the highest abundance is the sample’s predominant lineage, which can be searched for.

Errors (in steps 1 and 2) block a sample from being released until it is corrected. Warnings (in steps 3 and 6) are shown to the submitter before approval.

When the lineage barcodes are updated, all samples are processed again.

The dashboard can also estimate lineage prevalence over time with LolliPop (“LolliPop (time series)”). Unlike Freyja, LolliPop uses all samples of a site together: it fits the selected lineages’ mutations from Freyja’s barcodes to the SNV frequencies of the samples, weighting neighbouring dates with a Gaussian kernel, and reports 95 % confidence intervals from bootstrapping the mutations. Its parameters can be changed under “Advanced settings”.