EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Identifying multi-omics spatial domains via masked autoencoder and diffusion-based latent regularization with SpaMEDM

Chang Zhou iD, Mingxin Lu, Yilei Meng, Sangyu Hu, Yuping Zhu, Yizhuo Sun, Jianing Gu, Zhaoxuan Tian, Fei Yang iD, Jun Wang, Li Li iD

Briefings in Bioinformatics · 02.11.2026 · 27 (6)

DOI: 10.1093/bib/bbag520 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Briefings in Bioinformatics, online erschienen am 05.11.2026. Verfasst von Chang Zhou und 10 weiteren Autor:innen. Beteiligte Einrichtung: Zhabei Central Hospital of Jing'an District, School of Medicine, Tongji University, 619 Zhonghuaxin Road, Jing'an District 200070, Shanghai, China. Hauptthemen laut MEDLINE: Computational Biology. Die Arbeit nennt 53 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: National Natural Science Foundation of China, the Shanghai Committee of Science and Technology, the Shanghai Jing'an District Health Commission Key Discipline Project.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Spatial multi-omics data provide complementary molecular views of tissue organization. However, their integration remains challenging due to data sparsity, technical noise, and modality-specific distributions. Here, we present Spatial Masked autoEncoder and Diffusion-based Model (SpaMEDM), an end-to-end graph representation learning framework that augments a dual-attention graph-integration backbone with masked reconstruction and diffusion-based latent noise-prediction regularization. In the masked track, randomly masked spot features are recovered by leveraging modality- and graph-specific neighborhood information. In parallel, Gaussian perturbations are applied to modality-specific latent representations at randomly sampled noise levels, and a spatial-graph-informed network is trained to predict the injected noise. This diffusion-based objective is used exclusively during training to encourage robustness to latent perturbations, rather than to perform iterative reverse diffusion or sampling during inference. Furthermore, cosine-based paired-spot directional alignment and adjacency reconstruction encourage compatible cross-modality representations and preservation of the selected graph structure. SpaMEDM flexibly supports purely spatial, dual feature–spatial, and modality-specific feature-pruned spatial graph configurations. Evaluations across diverse spatial multi-omics datasets demonstrate competitive spatial-domain identification performance and produce integrated embeddings that facilitate downstream exploratory analyses.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Computational BiologyAutoencoderHumansMultiomicsAlgorithmsRepresentation Machine Learning

Stichwörter der Autor:innen

deep learningdiffusion-based regularizationmasked autoencoderspatial domain identificationspatial multi-omics

Autor:innen und Institutionen

  • Zhabei Central Hospital of Jing'an District, School of Medicine, Tongji University, 619 Zhonghuaxin Road, Jing'an District 200070, Shanghai, China.
  • State Key Laboratory of Cardiology and Medical Innovation Center, Shanghai East Hospital, School of Medicine, Tongji University, 150 Jimo Road, Pudong New Area District 200120, Shanghai, China.
  • Stem Cell Research Center, Medical School, Tongji University, 500 Zhennan Road, Putuo District 200120, Shanghai, China.
  • School of Biomedical Engineering, Anhui Medical University, Tanghe Road, Hefei 230601, Anhui, China.
  • Department of Cell and Genetics, Tongji University School of Medicine, 500 Zhennan Road, Putuo District 200120, Shanghai, China.

ORCID-Profile: Chang Zhou, Fei Yang, Li Li

Förderung

  • National Natural Science Foundation of China
  • the Shanghai Committee of Science and Technology
  • the Shanghai Jing'an District Health Commission Key Discipline Project
  • Shanghai Key Laboratory of Gene Editing and Cell Therapy for Rare Diseases
  • the Shanghai Jing’an District Health Commission Key Discipline Project

Zitieren

APA

Zhou, C., Lu, M., Meng, Y., Hu, S., Zhu, Y., Sun, Y., Gu, J., Tian, Z., Yang, F., Wang, J., & Li, L. (2026). Identifying multi-omics spatial domains via masked autoencoder and diffusion-based latent regularization with SpaMEDM. Briefings in Bioinformatics, 27(6). https://doi.org/10.1093/bib/bbag520

Vancouver

Zhou C, Lu M, Meng Y, Hu S, Zhu Y, Sun Y, et al. Identifying multi-omics spatial domains via masked autoencoder and diffusion-based latent regularization with SpaMEDM. Briefings in Bioinformatics. 2026;27(6). doi:10.1093/bib/bbag520

BibTeX
@article{Zhou2026,
  author = {Zhou, Chang and Lu, Mingxin and Meng, Yilei and Hu, Sangyu and Zhu, Yuping and Sun, Yizhuo and Gu, Jianing and Tian, Zhaoxuan and Yang, Fei and Wang, Jun and Li, Li},
  title = {Identifying multi-omics spatial domains via masked autoencoder and diffusion-based latent regularization with SpaMEDM},
  journal = {Briefings in Bioinformatics},
  year = {2026},
  volume = {27},
  number = {6},
  doi = {10.1093/bib/bbag520}
}
Kostenlos starten →