> And various dedicated AI models all made poor predictions of the results of that experiment, which casts doubts about the value of the AlphaGenome predictive map

I would not group AlphaGenome into the pile of failed predictions of other models. AlphaGenome deserves to get evaluated based off its own merits.

Yep. Sequence-to-function models are still very limited. AlphaGenome Atlas, despite the flashy branding, is unlikely to provide significant benefit to researchers.