possibility of VEP on long structral variation and repeats

[WeiCSong]

Hi, congratulations on the great tool! I'm wondering if EVO can be used to predict the effect of long variation like large deletion or repeat expansion, since it works well on long input. In such cases, the sequence for alt and ref would have very different lengths. Does this impact the likelihood score and require additional normalization? I think one workaround for the varying sequence length is to generate sequences from healthy cohorts of the same position, each carrying likely neutral variants and are of different length, and use their likelihood score to form a null distribution. Does this approach break any assumption of the EVO2 model? Thanks for your help!

[garykbrixi]

Hi @WeiCSong . Thanks for your question and interest. We hope Evo 2 can help with study of long sequence variation. In our preprint we have not yet assessed Evo 2 on disease variants affecting more than 64bp

[LauferVA]

I noticed this. It seems like it would be a major emphasis though - that is, why go to the trouble of making such an incredible context window if not to exploit it for things like SV

Will be interested to read this