Predicting single-molecule chromatin distribution from DNA sequence
- 1University Of Chicago
Abstract
Genome regulation unfolds through dynamic and stochastic single-molecule events. Deep learning models that read gene regulatory activities from DNA sequences have markedly advanced our ability to predict cell-population-level averages. However, they cannot capture the full spectrum of heterogeneity and dynamics at the single-molecule level, which directly reflect the underlying mechanisms of genome regulation. Here, we introduce a sequence-to-distribution (S2d) framework that directly models the entire probabilistic distribution of single-molecule chromatin states from DNA sequence alone. We used this framework to train Mola, a model of single-molecule chromatin accessibility. We harness Mola to investigate sequence-based regulatory code and mechanisms for multiple aspects of chromatin regulation at the single-molecule level. By quantifying sequence-driven nucleosome regulation, we reveal distinct modes of transcription factor engagement with nucleosomes, including a novel nucleosome-trapping factor, ZNF146. We characterized the sequence determinants governing the +1 nucleosome after transcription start sites and identified YY1 as the strongest +1 nucleosome-positioning factor. Leveraging long-context single-molecule modeling, we further discovered a shared sequence basis for the co-accessibility of regulatory elements and the co-footprinting of transcription factors. Joint prediction of DNA methylation and accessibility reveals a sequence-driven bimodal state mediated by methylation-sensitive transcription factors. Beyond static single-molecule snapshots, the S2d framework can generate pseudo-dynamic trajectories of chromatin states and propose a roadmap toward genome-wide chromatin dynamics. Moreover, we showed that Mola can predict and interpret the effects of genetic variants at the single-molecule level, revealing how sequence changes reshape distal regulatory coordination. Together, this sequence-to-distribution framework lays the foundation for understanding how genomic sequence encodes stochastic and dynamic genome regulation, and how genetic variation changes it, one molecule at a time.
Related
- DUET: sequence-only prediction of RNA-small molecule binding sites
- G4SCOPE identifies sequence features and epigenetic contexts of chromatin G-quadruplexes
- PathFold: Predicting the Entire Protein Folding Pathway from Protein Sequence Alone
- Adaptive-optics lattice light-sheet microscopy enables correlative single-molecule imaging of endogenous proteins in native environments
- Predicting Protein-Protein Interactions Using Vectorized Sequence Descriptors and Machine Learning Classifiers
BibTeX
@misc{wang2026predicting,
title = {{Predicting single-molecule chromatin distribution from DNA sequence}},
author = {R. Wang and J. Jin and S. Pott and J. Zhou},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.29.762992},
url = {https://www.biorxiv.org/content/10.64898/2026.10.29.762992v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Predicting single-molecule chromatin distribution from DNA sequence
AU - R. Wang
AU - J. Jin
AU - S. Pott
AU - J. Zhou
PY - 2026
DA - 2026/10/29
PB - bioRxiv
DO - 10.64898/2026.10.29.762992
UR - https://www.biorxiv.org/content/10.64898/2026.10.29.762992v1
AB - Genome regulation unfolds through dynamic and stochastic single-molecule events. Deep learning models that read gene regulatory activities from DNA sequences have markedly advanced our ability to predict cell-population-level averages. However, they cannot capture the full spectrum of heterogeneity and dynamics at the single-molecule level, which directly reflect the underlying mechanisms of genome regulation. Here, we introduce a sequence-to-distribution (S2d) framework that directly models the entire probabilistic distribution of single-molecule chromatin states from DNA sequence alone. We used this framework to train Mola, a model of single-molecule chromatin accessibility. We harness Mola to investigate sequence-based regulatory code and mechanisms for multiple aspects of chromatin regulation at the single-molecule level. By quantifying sequence-driven nucleosome regulation, we reveal distinct modes of transcription factor engagement with nucleosomes, including a novel nucleosome-trapping factor, ZNF146. We characterized the sequence determinants governing the +1 nucleosome after transcription start sites and identified YY1 as the strongest +1 nucleosome-positioning factor. Leveraging long-context single-molecule modeling, we further discovered a shared sequence basis for the co-accessibility of regulatory elements and the co-footprinting of transcription factors. Joint prediction of DNA methylation and accessibility reveals a sequence-driven bimodal state mediated by methylation-sensitive transcription factors. Beyond static single-molecule snapshots, the S2d framework can generate pseudo-dynamic trajectories of chromatin states and propose a roadmap toward genome-wide chromatin dynamics. Moreover, we showed that Mola can predict and interpret the effects of genetic variants at the single-molecule level, revealing how sequence changes reshape distal regulatory coordination. Together, this sequence-to-distribution framework lays the foundation for understanding how genomic sequence encodes stochastic and dynamic genome regulation, and how genetic variation changes it, one molecule at a time.
ER -