Reading more of a DNA change's consequences

The human problem

Most observed human genetic variation is non-coding, and interpreting its possible biological effects remains difficult.

The prior constraint

Existing sequence-to-function methods traded off input sequence length against prediction resolution and often covered only selected biological modalities.

AI’s actual role

A deep-learning model jointly predicts many molecular genomic measurements and scores the likely effect of a sequence variant across those modalities.

The documented result

In the paper's external variant-effect evaluations, AlphaGenome matched or exceeded the strongest available external models in 25 of 26 evaluations; it also recapitulated mechanisms of clinically relevant variants near the TAL1 oncogene.

Why it may matter

Researchers may be able to generate more integrated hypotheses about how a DNA variant changes gene regulation.

Limitations

The model predicts molecular effects; it does not diagnose a patient, establish causal disease mechanisms, or prove a treatment works.

Training and evaluation draw on existing human and mouse experimental datasets.

The paper describes non-commercial API access, not universal clinical availability.

Unresolved questions

Which predictions validate in disease-relevant tissues and individuals?

Can use of the model improve rare-disease interpretation or therapeutic development prospectively?