Preprint Finds AlphaGenome Systematically Underestimates Causal Variant Effects, Limiting Fine-Mapping
anshulkundaje · x · 2026-09-12
A new preprint by Drusinsky and Pollard (Gladstone Institutes) tests whether sequence-to-function (S2F) models like AlphaGenome can identify causal expression-modifying variants.
Key findings:
- S2F models classify putatively causal eQTL SNVs reasonably well, but dramatically underperform linear baselines at ranking individuals' gene expression from whole-genome sequences
- Systematic comparison against fine-mapped eQTLs shows AlphaGenome pervasively underestimates the effects of most causal variants and thus fails to fine-map most loci
- Widely cited issues — misdirected variant effect predictions and negative cross-individual correlations — are not egregious errors; the overlooked main driver is causal variant underestimation
Conclusion: causal variant underestimation is a major, overlooked cause of S2F underperformance, directly questioning their fine-mapping utility.
More from Research
- ParetoRSI-Gym benchmarks whether AI agents can recursively self-improve LLMs across domains — Haojin Wang · 2026-10-28
- Stanford's Ψ-0.6 is a unified vision-language world model with adaptive reasoning streams — Stanford · 2026-10-28
- Shanghai AI Lab's OpenESM shows energy-steered models can scale as foundation learners — Shanghai-AI-Laboratory · 2026-10-28
- CellTrack-Bench tests VLMs on microscopic cell tracking, exposing reasoning gaps — biohub · 2026-10-28
- Hybrid LLM agent scores 83.2%, beats nine VLA policies on new SO-101 robot arm benchmark — Truman Hickok · 2026-10-28
- DeepMind researcher: all of mathematics will soon be formalizable, with agents hunting proofs — ChrSzegedy · 2026-10-28