Skip to content
DraconiticResearch
Search Sources API
203,234 studies31 sourcesUpdated 7 min ago
Peer-reviewedCC BY 4.0Genome Biology

CellPatch: a flexible and efficient framework for single-cell foundation models empowered by heuristic gene patching

  • Hanwen Zhu
  • Yushun Yuan
  • Jiyuan Yang
  • Kangwen Cai
  • Nana Wei
  • Senxin Zhang
  • Lu Wang
  • Wen-Jie Jiang
  • Yuanchen Sun
  • An Liu
  • Futing Lai
  • Yu-Juan Wang
3 more authors
  • Zeyu Ma
  • Xiaoqi Zheng
  • Hua-Jun Wu
Open at sourceDOICite

Abstract

Foundation models have advanced single-cell omics analysis but often remain too complex for practical use. We present CellPatch, a lightweight foundation model combining cross-attention with heuristic gene patching to reduce computational complexity during pretraining and fine-tuning while enabling inspection of gene-attention patterns. Across single-cell RNA-sequencing datasets spanning multiple organs and tissue states, CellPatch achieves state-of-the-art performance on diverse prediction tasks. Its modular design supports linear, adapter-based and task-specific architectures, enabling CellPatch to be adapted to diverse biological tasks.

Related

  1. OpenDVP: an experimental and computational framework for community-empowered deep visual proteomics
    Jose Nimo, Sonja Fritzsche, Daniela S. Valdes +9
    Nature MethodsPeer-reviewedCC BY 4.0Full text6 citedMolecular biology & gene regulationPMC
  2. Time-lapse RESOLFT nanoscopy empowered by deep learning
    Guillaume Minet, Anirban Ray, Francesca Pennacchietti +3
    Nature CommunicationsPeer-reviewedCC BY 4.0Full textBiophysicsCrossref
  3. [Iron porphyrin metal-organic framework material for efficient enrichment of earthworm polypeptides].
    Kun-Song Zhan, Ming-Zhu Tang, Ying Lin +2
    Se pu = Chinese journal of chromatographyPeer-reviewedCC BYFull textMolecular biology & gene regulationEurope PMC
  4. New heuristic method for cancer classification using gene expression data.
    Shengyu Gu, Tamara Savina, Dimitar Monov +1
    Cancer imaging : the official publication of the International Cancer Imaging SocietyPeer-reviewedCC BY-NC-NDFull textMolecular biology & gene regulationEurope PMCPMC
  5. BulkFormer: A large-scale foundation model for bulk transcriptomes
    B. S. Kang, Rui Fan, Meizheng Yi +2
    Cell SystemsPeer-reviewedLicence not statedFull text17 citedMolecular biology & gene regulationOpenAlex

Cite

BibTeXRIS
BibTeX
@article{zhu2026cellpatch,
  title = {{CellPatch: a flexible and efficient framework for single-cell foundation models empowered by heuristic gene patching}},
  author = {Hanwen Zhu and Yushun Yuan and Jiyuan Yang and Kangwen Cai and Nana Wei and Senxin Zhang and Lu Wang and Wen-Jie Jiang and Yuanchen Sun and An Liu and Futing Lai and Yu-Juan Wang and Zeyu Ma and Xiaoqi Zheng and Hua-Jun Wu},
  journal = {Genome Biology},
  year = {2026},
  month = oct,
  doi = {10.1186/s13059-026-04329-y},
  url = {https://link.springer.com/article/10.1186/s13059-026-04329-y},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - CellPatch: a flexible and efficient framework for single-cell foundation models empowered by heuristic gene patching
AU  - Hanwen Zhu
AU  - Yushun Yuan
AU  - Jiyuan Yang
AU  - Kangwen Cai
AU  - Nana Wei
AU  - Senxin Zhang
AU  - Lu Wang
AU  - Wen-Jie Jiang
AU  - Yuanchen Sun
AU  - An Liu
AU  - Futing Lai
AU  - Yu-Juan Wang
AU  - Zeyu Ma
AU  - Xiaoqi Zheng
AU  - Hua-Jun Wu
PY  - 2026
DA  - 2026/10/30
JO  - Genome Biology
DO  - 10.1186/s13059-026-04329-y
UR  - https://link.springer.com/article/10.1186/s13059-026-04329-y
AB  - Foundation models have advanced single-cell omics analysis but often remain too complex for practical use. We present CellPatch, a lightweight foundation model combining cross-attention with heuristic gene patching to reduce computational complexity during pretraining and fine-tuning while enabling inspection of gene-attention patterns. Across single-cell RNA-sequencing datasets spanning multiple organs and tissue states, CellPatch achieves state-of-the-art performance on diverse prediction tasks. Its modular design supports linear, adapter-based and task-specific architectures, enabling CellPatch to be adapted to diverse biological tasks.
SN  - 1474-760X
ER  -