Could AlphaGenome help interpret SMN2 splicing variants?
AlphaGenome predicts regulatory and splicing effects from DNA sequence. For SMA, that makes it an interesting complement to sequence-space studies of SMN2 exon 7 — but not a substitute for drug-response or experimental data.
Why the connection is interesting
GAYAR's current SMN2 work studies how sequence variation around exon 7 relates to splicing behavior and response patterns under risdiplam and branaplam. AlphaGenome approaches a neighboring question from another direction: given DNA sequence, what regulatory and splicing consequences does the model predict?
Google DeepMind's AlphaGenome takes up to one megabase of DNA sequence and predicts many functional genomic signals, including splice-site usage and splice-junction coordinates and strength. Its newer AlphaGenome Atlas pre-computes variant-effect predictions for single-nucleotide changes across the human genome. That makes the system particularly relevant to questions where sequence changes may alter exon inclusion or nearby regulatory behavior.
Where it could help
For SMN2, AlphaGenome could act as an independent sequence-to-function prior: a way to ask whether variants that occupy nearby regions of sequence space are also predicted to have similar splicing consequences. That could complement representation-focused work such as ML4Spatial, which asks what biological structure is preserved when an engineered SMN2 variant space is compressed into a two-dimensional representation.
The distinction matters. AlphaGenome is not a model of risdiplam or branaplam response, and the Atlas is specifically a catalogue of single-nucleotide substitutions. Engineered variant libraries can contain sequence combinations outside that Atlas scope, where the broader model or API would be the more relevant tool.
What it cannot establish
A model prediction is a hypothesis, not validation. AlphaGenome cannot by itself show that a predicted splicing change occurs experimentally, that it causes disease severity, or that a splice-modifying compound will produce a particular response. Those questions still require assay data and careful experimental design.
Sources: Google DeepMind — AlphaGenome, AlphaGenome Atlas, and the 2026 Nature paper.