Predicting nucleotide change into ambiguous nucleotide with AlphaGenome

Help & Support
1

Thank you for your very nice work!

I was wondering how does AlphaGenome handle the input of ambiguous “N” as the variant alternative base in variant specification. It doesn’t return an error and it returns a different output (different quantile scores and different output dimensions) in the following input versions for the following tracks.

SPLICE_SITES, SPLICE_SITE_USAGE, SPLICE_JUNCTIONS

variant_chromosome = ‘chr14’

variant_position = 93686572

REF = ‘T’, ALT = ‘A’, 4039 rows

REF = ’T’, ALT = ’N’, 2204 rows

REF = ‘G’, ALT = ‘N’, 4773 rows

REF = ’A’, ALT = ’N’, 10278 rows

REF = ’C’, ALT = ’N’, 4406 rows

2

Hi @alptaciroglu, welcome to the forum!

We don’t do anything special for alternate alleles, so if you pass N for an alt base, then this ends up being 1-hot encoded into [0, 0, 0, 0] and passed to the model :smiley:

Hope that answers your question!

1 Like