CNV input

[StephanHolgerD]

Hi, thanks for providing talos !

Is it possible to include CNV data in to the analyses as well ?

best,
Stephan

[MattWellie]

Hi again @StephanHolgerD,

Talos currently works with CNV data, but in a very specific way. We have two types of callset, each generated and annotated by a GATK tool, though I appreciate other users will have data from different sources.

We're using two separate SV pipelines internally, one is GATK-gCNV for exomes, the other is a multimodal GATK-SV for genomes. Talos works with VCFs generated by both of these tools, but we've not had data in other formats to check against.

The consequence annotation for both types of variant data is done by GATK-SVAnnotate, and external population frequency is applied by a rather complicated series of steps in the GATK-SV pipeline. As consequence and frequency annotations are integral to Talos' filtering and prioritisation, we need incoming SV data to have the expected annotation fields.

With Talos we've tried to maintain a uniform entrypoint - instead of sites all bringing their own annotated data, and Talos having to accommodate a range of annotation tools and formats, Talos has taken ownership of the variant annotation pipeline. This is a little wasteful for sites with pre-annotated data, but makes deployment much simpler.

The logical step here is for Talos to implement a SV annotation step into the workflow, so any site with small and structural variant data can integrate the two. We've not taken this step yet, but can try to evaluate/prioritise that if you have data you want to analyse.

Another factor is that because the CNV/SV annotation workflow steps don't exist, the CNV/SV input path hasn't been added to the NextFlow implementation. This will also need to be added to make that functionality available to others.

[StephanHolgerD]

Hello, thanks for the explanation.

An entrypoint for SV variants would be great.

We are planing to set up a pipeline to run a periodically re-evaluation of unsolved datasets at our institute.

At the moment we are doing some test runs to see if talos is an option for us.

[MattWellie]

Hi @StephanHolgerD, apologies for leaving this hanging. We'd like to pick this thread back up and get it working for you.

There are 3 factors in our current SV implementation:

1. VCF must be standard form/specification compliant
2. Annotated with GATK SVAnnotate
3. Annotated with population/callset frequencies

For small variants we have taken ownership of the entire annotation process, as that was easier for us to handle. For SVs there is such a wide range of callers and technologies that it's difficult for us to own that process.

I plan to do the following:

1. Expose a SV data entrypoint in the NextFlow workflow
2. Evaluate the filtering benefit from population/callset frequency annotations, decide whether they can be made optional (at least initially) - this would make GATK SVAnnotate annotations the only essential component to ingest SV data.

Are you able to run your CNV data through SVAnnotate to be compliant with the processing model we already have in Talos? Or if you prefer a different annotator could you let us know which that is, and we can see if it's simple enough to adapt to use that?