DNA language models and AlphaGenome predict different things

A DNA change can affect how a gene works, but different genomic models give you different kinds of evidence about it. A DNA language model learns patterns in DNA sequences. An autoregressive model such as Evo 2 can compare how likely the original and changed sequences are. That score does not directly tell you how much gene expression will change. AlphaGenome instead learns from experimental measurements and predicts readouts such as gene expression and splicing. Comparing its predictions for the original and changed DNA can suggest which molecular processes are affected. The scoring method depends on the readout; it is not always simple subtraction. Neither kind of prediction establishes a diagnosis. AlphaGenome's developers explicitly restrict its predictions to research use. Developer documentation: https://github.com/google-deepmind/alphagenome Guide and diagram: https://rewirebio.io/blog/genomic-foundation-models-in-the-diagnostic-lab/ Disclosure: rewirebio.io is my site. The guide, this post and the explanatory diagram are AI-assisted. https://preview.redd.it/lgmcxpl4yewh1.png?width=1520&format=png&auto=webp&s=dd634ff54e726f67e7b457c834e9bd60656bdb1c