How to use a different GENCODE version (e.g., v49) with score_variant()?

Help & Support
1

Hello all,

I am trying to calculate variant scores related to RNA-seq and splicing (splice sites, splice site usage, splice junctions).

It appears that AlphaGenome’s score_variant() function has GENCODE v46 built in by default. However, I would like to use the latest version, GENCODE v49.

I have a few questions regarding this:

  1. Is there a way to change the GENCODE version in the score_variant() function?

  2. If not, should I generate a .gtf.gz.feather file from the v49 GTF file using the script below and then calculate variant scores directly using the predict_variant() function?
    alphagenome/scripts/process_gtf.py at main · google-deepmind/alphagenome · GitHub

If I need to use predict_variant(), is there a formula or example code for calculating each score?

In particular, according to the documentation, the splice_junctions score is calculated as max(|log(ALT) - log(REF)|) from “top-k splice sites” within the gene body:

  • What is the value of k?
  • What criteria are used to select the “top-k” splice sites?

Reference: How variant scoring works — AlphaGenome

Thank you in advance for your help.

How to get signed (non-absolute) raw scores for splicing variants?
2

Hello,

Apologies for not replying sooner. You should now be able to run the model locally with an updated GTF by using our code + weights.

This should also help clarify your queries re: junction scoring (top-k is 256 for the scorer).

Thanks!

Tom