I wanted to ask if, instead of introducing single-nucleotide variations and checking their effects on the signal using dna_model.predict_variant() — which has been very useful — does AlphaGenome also support mutating entire segments of nucleotides, it could be one location or several locations?
For example, is it possible to modify a 20 bp sequence, at one location or potentially at multiple locations, and then analyse the resulting impact on the signal? If so, how can this be done?
It might be possible. If you construct the reference allele and alternative allele to be the same length, but include the multiple SNPs you’d like to predict. e.g. ATTGCA>GTTACT, which include 3 SNPs. Please feel free to share any feedback if this works for you.
If an insertion or deletion results in different sequence lengths, is it necessary to pad with 'N’s beforehand? Or does Alphagenome have other handling logic, particularly during visualization?