Predicting cell-state-specific gene regulation from DNA across species

Preprint (not peer-reviewed): Liu et al. (corresponding institution Arc Institute; senior author Greenberg ME) present Bolero, a sequence-to-function model whose parameters condition on atlas-scale cell-state embeddings and other biological variables to predict cell-state-specific chromatin accessibility and transcript abundance from DNA. Trained on 10.7 million cells from 33 datasets across six mammals, it is reported to match or exceed specialised models; the authors report that extrapolating cell-state embeddings extended predictions to cell states absent from training, that an AP-1 activity score as a conditioning axis predicted activity-regulated chromatin changes without stimulation data, and that it predicted cell-type-specific accessibility in species excluded from training. Self-reported; no independent evaluation yet.

Authors / organisation

Liu H, Mao S, Gallent T, Chen Y, Yan Z, Vogel I, Traunmuller L, Zhang R, Buenrostro JD, Zhou J, Greenberg ME.

Venue

bioRxiv (preprint, not peer-reviewed)

Date

2026-10-20

Type

preprint

Year

2026

DOI

10.64898/2026.10.19.760026

Verification

bioRxiv API details (title, 11 authors, posted 2026-10-20, v1, new results, genomics, CC BY-NC, corresponding institution Arc Institute, abstract) + Crossref posted-content record; doi.org resolves to biorxiv.org (10.64898 is bioRxiv's current DOI prefix) fetched 2026-10-21; full text not read

Foundation modelGene regulatory networkCross-speciesGenerative AI & diffusion

Open in the searchable library → · JSON