Genome annotations for segmented CDSs etc

[jameshadfield]

Auspice will soon be able to parse an extended version of the genome_annotation which will allow segmented CDSs, wrapping CDSs and extra metadata. We need a way to export this information from Augur.

Schema changes

general

The nuc strand cannot be "-" (-ve strand genomes are represented as their reverse complement). This isn't a change to Auspice's behavior but is now enforced. The strand is optional for nuc.

Each key/object pair in genome_annotations now corresponds to a CDS rather than a gene and our language (help / schema etc) should be updated accordingly.

CDS length is now verified to be a multiple of 3 (within Auspice) and if it's not the CDS is not displayed.

segmented CDS

The start and end properties may be omitted and replaced with an array of segments, each with 1-based (GFF) coordinates:

segments?: {start: number, end: number, name?: string}[]

The order of segments is important and corresponds to the order the respective translations appear in the protein sequence. Note that start<end always, even if the CDS is on the negative strand.

(The name for an individual segment may or may not be part of the schema, I will update this once we've finalised it in Auspice.)

wrapping CDS

A wrapping CDS may be expressed via segments (as above) or by specifying an end coordinate beyond the length of the genome, following GFF format.

(optional) metadata

gene?: string: Displayed in the on-hover tooltip in Auspice. If multiple CDSs have the same gene then they will be given the same colour by Auspice.

color?: string: User specified colour for the CDS. The value must be a CSS colour string or a colour hex.

display_name?: string: A more verbose name for the CDS, shown in the on-hover info box.

description?: string: Shown in the on-hover info box.

(future) within CDS annotations

This isn't yet in Auspice, but will be, so it is important context to consider when implementing the Augur side of things. The syntax may change slightly.

features?: {
  name: string;
  description: string;
  segments: {start: number, end: number}[];
}[]

Current augur workflow

(Remember that augur currently can't handle complex CDSs)

1. augur ancestral infers nuclotide sequences and adds the "nuc" annotation to the resulting node-data JSON
2. augur translate uses a GenBank / GFF file to translate simple CDSs and creates per-CDS (per-gene) annotations in the resulting node-data JSON
3. augur export simply passes this annotations block through to the final dataset JSON

Future augur workflow

I believe the typical augur workflow, especially for complex CDSs will be:

1. Generate per-CDS translations for each sample via Nextclade, which uses (exclusively?) GFF annotations
2. Use augur ancestral to infer the translated sequences, per CDS, for internal nodes (implemented via this PR). This is probably the place to also generate the annotations block, which means that augur ancestral will need to parse the GFF file initially used by Nextclade.
3. augur export simply passes this annotations block through to the final JSON, as it currently does.

I'm not sure what this means for the future of augur translate, but the approach we use within augur ancestral to parse the GFF (or GenBank?) file and create the resulting JSON annotations will be able to be used by augur translate if we wish to do so in the future.

There is still the question of how to export optional metadata for each CDS, such as "display name", "color" etc. For the time being I think it's ok to leave this as a script-based "optional extra" for workflows, but perhaps others see a nice way to implement this.

Related issues / discussions:

#953 covers GFF parsing in augur

We've discussed parsing GFFs within Augur recently in this slack thread

Auspice PR related to this work is here

Which pathogens does this affect?

For the time being, not many.

• nCoV - if we maintain the separate ORF1a and ORF1b for backwards compatibility, the only CDS that needs to utilise the new schema is RdRP (NSP12). We may also want to choose our own colours etc. This can easily be done by a small script in the pipeline.
• HepB - a prototype repo not yet in live production
• Ebola - not in current development or seeing new genomes, but we should add the slip site at some point.
• HIV - unsure of Richard's plans here

Path forward

This can be implemented in steps / multiple PRs

• Update the schema, which at the least would allow workflows to introduce the new annotations via scripts. I will try to do this shortly. This is done in Annotations schema updates #1281
• Parse GFF/GenBank (?) files and export them in this format
• Implement the within-CDS features later on once Auspice supports them?

[corneliusroemer]

Thanks for you work and thoughts here! Just came across this when trying to make a nice ORF1ab dataset for Nextclade v3. I noticed that augur translate wasn't correctly making use of the complex CDS features in the input gb file yet. I'll see if I can get augur translate to output complex cds correctly to the new schema.

[corneliusroemer]

@jameshadfield I think I've managed to make augur translate output the new complex CDS annotation - at least when one gives a genbank file as annotation in #1333.

Can you explain what you mean by "Implement the within-CDS features later on once Auspice supports them?"?

[jameshadfield]

I think I've managed to make augur translate output the new complex CDS annotation - at least when one gives a genbank file as annotation in #1333.

Nice!

Implement the within-CDS features later on once Auspice supports them?

The ability for annotations to define regions within a CDS, e.g. RBD in spike, epitope sites in HA, which Auspice can then show. Here's (slack) a prototype, but the UI hasn't been iterated on much. I don't think any schema's been decided on, but we can do that now if it's helpful?