AlphaGenome Atlas applies model predictions across the human genome to estimate how single-letter DNA changes could affect processes such as gene expression and RNA splicing. The resulting map covers roughly nine billion possible substitutions and is described as approximately one petabyte of precomputed data.
The AVI score condenses many predicted effects into a ranking signal so researchers can decide which variants deserve closer investigation. A high score is not treated as proof that a variant causes disease; it supplies a more specific hypothesisAI hypothesis generation uses AI to propose testable explanations or strategies from observations and data. about mechanisms such as disrupted gene activity or incorrect RNA splicing.
The transcript reports that researchers used the score to prioritize a previously overlooked DNM1 variant in a severe epilepsy case, with laboratory experiments supporting the predicted splicing effect. It also describes a study of more than 54,000 UK Biobank participants that reportedly found additional non-coding associations by grouping variants according to predicted molecular effects.
The larger argument is that widely accessible prediction maps can help research groups extract more value from existing genomic data and choose better experiments. The closing vision combines multiple specialized models into a more complete computational representation of human biology, while acknowledging that prediction, experimental validation, and treatment development remain distinct steps.
Watch on YouTube
