Diploid-aware genomic language model released: most models train on genomes that don't exist
OdedRechavi · x · 2026-09-15
A new diploid-aware genomic language model has been released. The authors point out a overlooked flaw in the field: human genomes are diploid (two alleles per locus), yet most genomic language models read a single haploid sequence — meaning the genomes those models are trained on don't actually exist. The new model is now available.
More from Research
- Research advice: imagine your line of work succeeding in 5 years to structure it today — abursuc · 2026-09-15
- RareGap Atlas maps 11,460 rare diseases into 4,073 queryable intervention hypotheses — RichardsonDx · 2026-09-15
- Princeton eval finds reasoning models fail structurally equivalent task variants, lacking systematicity — princetonu · 2026-09-15
- TREVIS learns sparse decision trees via transformer variational auto-encoders — Giacomo Fidone · 2026-09-15
- E2A-Bench benchmarks evidence-to-action reliability in financial chart reasoning, finds VLMs falter — Xiaoya Wang · 2026-09-15
- Reranking more docs backfires: Recall@10 peaks near 100, then declines, study finds — CShorten30 · 2026-09-15