Translate & Align V-Pipe Reads

[gordonkoehn]

Currently, we use Nextclade to translate from nucleotides to amino acids.

We also use it to realign the notes and amino acids, even though our nucleotides are already aligned.

This seemed fine for small test data, yet it may be infeasible or a mere waste of resources to realign the reads again.

So consider rewriting the translate functions.

[gordonkoehn]

Potential Solution: call functions within Nextclade are told after the alignment.

• Reach out the the nextclade devs

[gordonkoehn]

See their response:

• `.sam` to amino acids – can I just translate ? nextstrain/nextclade#1556

In particular from rneher

A simple script could translate those already aligned reads. You'd need to figure which ORFs you read falls into, the reading frame, and then use something like Bio.Sequence.translate from the biopython package to translate the sequence.

[gordonkoehn]

So, we probably need a custom tool, yet I am unsure how to get the new amino acid positions.

Plus in general I don't know how to deal with the insertions

• Read Pairing & Merge #54

Can I just translate the nucleotide insertion to amino-acid insertions?

[gordonkoehn]

@DrYak, for now, I am ignoring insertions completely. There are too many biological unknowns for me here. I'd cherish your advice here sometime.

[gordonkoehn]

The issue:

[gordonkoehn]

Need some input before I can continue.

[gordonkoehn]

Take-aways from chat with @LaraFuhrmann

• BIG CAVEAT to uploading the .bam reads to SILO is that this discards all of V-Pipe's mutation calling efforts, so the database will contain sequencing artefacts.
• Yet, anyone will still be able to query, which is useful in urgent situations quickly.
• As a rough estimate, V-Pipe's computational effort is 15 % preprocessing / 25 % nucleotide alignment / 60 % mutational calling.
• The current plan hence is just to take all .bam take single reads to align with nextclade and once down, merge pair reads
• Basically, we only use V-Pipe's preprocessing and use Nextclade alignment / at the end, we again have V-Pipe's post-processing.

So actionable:

• implement to run nextclade for batches of the reads to still run in a small docker, just for longer.
• this means sr2silo is a computationally expensive step

@DrYak FYI. Let's also chat about this before I do it.

[gordonkoehn]

Take-aways from chat with @DrYak:

There exists, indeed, no code for this as of now. This is because most of the time, people will use mutations called on nucleotides and only translate these mutations, not entire alignments.

There are two options:

• 1. build it from the ground( with pysam and proper logic to handle all coding frames, handle corner cases)
• 2. hack Nextclade and hook into the process after the alignment they do

In both cases, the steps would be to take a .sam of the single reads and make a .sam with paired reads, i.e. using Micha's tool.

Once that exists, translate the amino acid alignment with either option.

Option 1) would probably take me 1-2 Months to handle properly – hard to estimate – for my lack of experience.

Option 2) could be quick and handle all corner cases.

[gordonkoehn]

This is a tool that Niko shared

• Check out this tool VIRULINGN

it appears it does the alignment itself. So this is not what we want, but it reads like there are some difficulties with the translation. Which just proved my point to use a well-supported tool. Then Nextclade will be their better choice if I hack something.

[gordonkoehn]

• How does V-Pipe align? Is that better in any way than Nextclade?

Ivan mentioned some probabilistic work in the alignment and theorised that Nextclade might do something more simplistic.