Refining sequence-to-activity models by increasing model resolution

Nuria Alina Chandra, Yan Hu, Jason D Buenrostro, Sara Mostafavi, Alexander Sasse · 2026

Read Refining sequence-to-activity models by increasing model resolution by Nuria Alina Chandra, Yan Hu, Jason D Buenrostro, Sara Mostafavi, Alexander Sasse (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13427759.txt.

Description

Chandra and colleagues upgraded in 2026 a deep learning model that reads DNA sequence to predict where the genome is open for business in immune cells. Open regions, called accessible chromatin, hold the switches that set cell identity. Their earlier model used only regional openness scores. The new one also learns the fine pattern of enzyme cuts within each region, down to single bases.

Trained at base-pair resolution, the model, called bpAI-TAC, predicted differences in openness between cell types better than the older version. Learning several cell types together beat training each alone. Comparing what the models paid attention to revealed new DNA motifs with strong effects that appear only when the fine cut pattern is included. The authors argue that resolution itself teaches the model a richer grammar of gene control.

Catalog record

CREATORNuria Alina Chandra, Yan Hu, Jason D Buenrostro, Sara Mostafavi, Alexander Sasse
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE40 KB
LICENSEhttps://creativecommons.org/licenses/by/4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/article/MED/PMC13427759

Browse the full library · About the Tower of Babel