AlphaGenome Enhances Personal Gene Expression Prediction but Retains Key Limitations

AlphaGenome significantly outperforms Enformer for personal, tissue-specific gene expression across 11,471 eligible gene-tissue pairs, winning 3,103 vs 828 pairs (win ratio 3.75).

• Why it matters: Sequence-to-function models are criticized for limited accuracy in predicting individual-level gene expression, leaving personal expression utility incompletely assessed.
• What they did: The study evaluated AlphaGenome’s sequence-based predictions for tissue-specific, personal gene expression across 1,000 genes and 11,471 gene-tissue pairs, using correlation comparisons and a global permutation test (P = 2.0e-4).
• The result: AlphaGenome’s advantage is supported by a median correlation difference of 0.046, but pretrained AlphaGenome substantially underperforms Elastic Net and Random Forest trained on individual-level data.