Skip to content
DraconiticResearch
Search Sources API
194,161 studies31 sourcesUpdated 45 min ago
Peer-reviewedCC BY 4.0PLOS Computational Biology

An extensible deep learning framework for multi-type chromatin interaction prediction through hierarchical adaptation strategies

  • Shuaibin Wang1
  • Tong Chen1
  • Weixu Liu2
  • Xuejie Zheng3
  • Zhongxin Yang1
  • Zhen Liang1,4
  • Yin Shen1,5
5 affiliations
  1. 1School of Biomedical Engineering, Anhui Medical University, Hefei, Anhui, China
  2. 2Department of Computer Science and Technology, Anhui Medical University, Hefei, Anhui, China
  3. 3Department of Pediatrics, The First Affiliated Hospital of Anhui Medical University, Hefei, Anhui, China
  4. 4The Affiliated Chuzhou Hospital of Anhui Medical University, The First People’s Hospital of Chuzhou, Chuzhou, Anhui, China
  5. 5Second Affiliated Hospital of Anhui Medical University, Hefei, Anhui, China
Read full textPDFDOICite

Abstract

Three-dimensional chromatin interactions play essential roles in gene regulation, yet computational prediction of diverse interaction types remains largely dependent on specialized, single-task models. Here, we present UniLoopNet, a multi-task deep learning framework that predicts multiple types of chromatin interactions including enhancer–promoter interactions, enhancer–enhancer interactions, CTCF-mediated loops, and stripe-associated interactions from DNA sequences. UniLoopNet employs a shared encoder–decoder architecture with hierarchical adaptation strategies, integrating multi-task, transfer, and continual learning based on task semantic similarity. Comprehensive benchmarking across multiple cell lines demonstrates that UniLoopNet outperforms existing specialized and multi-task approaches, with AUROCs exceeding 0.90 in most tasks. Under stringent evaluation settings, including chromosome-matched, distance-matched, realistic imbalanced, and locus-disjoint tests, AUROCs range from 0.8934 to 0.9366. In the realistic 1:5 class-imbalance setting, the AUPRCs for enhancer–promoter and enhancer–enhancer interaction prediction are 0.7956 and 0.6935, respectively. Cross-cell-line transfer experiments further show that applying UniLoopNet to another cell line requires adaptation by supervised fine-tuning on target-cell-line data, after which performance is comparable to that of within-cell-line training. Notably, the framework exhibits data efficiency in experiments with substantially reduced training data. Interpretability analyses reveal that the learned sequence features are associated with known transcription factor binding sites and uncover differential regulatory effects of disease-associated variants. UniLoopNet provides a versatile and data-efficient framework for sequence-based prediction and interpretation of diverse 3D genome interactions, complementing existing sequence-to-map approaches while improving the understanding of sequence-encoded regulatory potential. Author summary: The three-dimensional organization of the genome plays a critical role in regulating gene expression by bringing distant regulatory regions into physical contact. However, current computational methods for predicting chromatin interactions are usually designed for individual interaction types, such as enhancer–promoter loops or protein-mediated contacts, requiring separate models for each task. This limits their ability to transfer knowledge across related genomic problems, especially when experimental data are limited. Here, we develop UniLoopNet, an extensible deep learning framework that can predict multiple types of chromatin interactions from DNA sequences. Instead of relying on a single learning strategy, UniLoopNet automatically adapts its knowledge-transfer strategy according to the similarity between different prediction tasks, enabling efficient learning of new interaction types. UniLoopNet achieves accurate predictions across diverse chromatin interaction tasks, is data-efficient, and identifies sequence patterns associated with different regulatory interactions. Applying UniLoopNet to disease-associated genetic variants further suggests how non-coding mutations may influence genome organization. Our study thus provides a flexible computational framework for studying genome regulation and interpreting functional genomic variation.

Related

  1. DeepEnzyKat: Turnover number prediction through enzyme-substrate interaction modeling and adaptive expert ensembles
    Guangran Zhang, Yujing Zhao, Yongwen Zhang +6
    ChemRxivPreprintCC BYMolecular biology & gene regulationChemRxiv
  2. Establishing standardized evaluation protocols for peptide-target interaction prediction: the PTI-TAPE consensus framework eDelphi
    William J. Waldock, Guangshun Wang, Ara Darzi +2
    Frontiers in bioinformaticsPeer-reviewedCC BYFull textMolecular biology & gene regulationEurope PMC
  3. A novel matrix multiplication framework for modeling genotype-by-environment interaction in genomic prediction
    Osval A. Montesinos-López, Abelardo Montesinos-López, Cricelio Montesinos-Lopez +5
    Frontiers in geneticsPeer-reviewedCC BYFull textGeneral biochemistryEurope PMC
  4. scChronos: hierarchical temporal context learning for missing snapshot prediction in single-cell time series
    Jiongsen Yao, Yunfei Zhang, Yong Xu +2
    BioinformaticsPeer-reviewedCC BY 4.0Full textMolecular biology & gene regulationPMC
  5. From biofilm tolerance to antimicrobial resistance: molecular adaptation strategies and emerging therapeutic vulnerabilities in resistant pathogens
    Kartik Chandra Guchhait, Subhamoy Dey, Sumita Rani Pradhan +2
    Frontiers in MicrobiologyPeer-reviewedCC BY 4.0Full textMolecular biology & gene regulationCrossref

Cite

BibTeXRIS
BibTeX
@article{wang2026extensible,
  title = {{An extensible deep learning framework for multi-type chromatin interaction prediction through hierarchical adaptation strategies}},
  author = {Shuaibin Wang and Tong Chen and Weixu Liu and Xuejie Zheng and Zhongxin Yang and Zhen Liang and Yin Shen},
  journal = {PLOS Computational Biology},
  year = {2026},
  month = oct,
  doi = {10.1371/journal.pcbi.1014911},
  url = {https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1014911},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - An extensible deep learning framework for multi-type chromatin interaction prediction through hierarchical adaptation strategies
AU  - Shuaibin Wang
AU  - Tong Chen
AU  - Weixu Liu
AU  - Xuejie Zheng
AU  - Zhongxin Yang
AU  - Zhen Liang
AU  - Yin Shen
PY  - 2026
DA  - 2026/10/27
JO  - PLOS Computational Biology
DO  - 10.1371/journal.pcbi.1014911
UR  - https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1014911
AB  - Three-dimensional chromatin interactions play essential roles in gene regulation, yet computational prediction of diverse interaction types remains largely dependent on specialized, single-task models. Here, we present UniLoopNet, a multi-task deep learning framework that predicts multiple types of chromatin interactions including enhancer–promoter interactions, enhancer–enhancer interactions, CTCF-mediated loops, and stripe-associated interactions from DNA sequences. UniLoopNet employs a shared encoder–decoder architecture with hierarchical adaptation strategies, integrating multi-task, transfer, and continual learning based on task semantic similarity. Comprehensive benchmarking across multiple cell lines demonstrates that UniLoopNet outperforms existing specialized and multi-task approaches, with AUROCs exceeding 0.90 in most tasks. Under stringent evaluation settings, including chromosome-matched, distance-matched, realistic imbalanced, and locus-disjoint tests, AUROCs range from 0.8934 to 0.9366. In the realistic 1:5 class-imbalance setting, the AUPRCs for enhancer–promoter and enhancer–enhancer interaction prediction are 0.7956 and 0.6935, respectively. Cross-cell-line transfer experiments further show that applying UniLoopNet to another cell line requires adaptation by supervised fine-tuning on target-cell-line data, after which performance is comparable to that of within-cell-line training. Notably, the framework exhibits data efficiency in experiments with substantially reduced training data. Interpretability analyses reveal that the learned sequence features are associated with known transcription factor binding sites and uncover differential regulatory effects of disease-associated variants. UniLoopNet provides a versatile and data-efficient framework for sequence-based prediction and interpretation of diverse 3D genome interactions, complementing existing sequence-to-map approaches while improving the understanding of sequence-encoded regulatory potential. Author summary: The three-dimensional organization of the genome plays a critical role in regulating gene expression by bringing distant regulatory regions into physical contact. However, current computational methods for predicting chromatin interactions are usually designed for individual interaction types, such as enhancer–promoter loops or protein-mediated contacts, requiring separate models for each task. This limits their ability to transfer knowledge across related genomic problems, especially when experimental data are limited. Here, we develop UniLoopNet, an extensible deep learning framework that can predict multiple types of chromatin interactions from DNA sequences. Instead of relying on a single learning strategy, UniLoopNet automatically adapts its knowledge-transfer strategy according to the similarity between different prediction tasks, enabling efficient learning of new interaction types. UniLoopNet achieves accurate predictions across diverse chromatin interaction tasks, is data-efficient, and identifies sequence patterns associated with different regulatory interactions. Applying UniLoopNet to disease-associated genetic variants further suggests how non-coding mutations may influence genome organization. Our study thus provides a flexible computational framework for studying genome regulation and interpreting functional genomic variation.
SN  - 1553-7358
SN  - 1553-734X
ER  -