scYOU: a multi-scale yoked optimization and unsupervised framework for analysis of cellular heterogeneity and complex-associated protein module recovery
Briefings in Bioinformatics · 02.11.2026 · 27 (6)
DOI: 10.1093/bib/bbag626 ↗Kurz gefasst
Fachartikel in der Zeitschrift Briefings in Bioinformatics, online erschienen am 09.11.2026. Verfasst von Yuan Gan und 2 weiteren Autor:innen. Beteiligte Einrichtung: School of Mathematics, Renmin University of China, No. 59 Zhongguancun Street, Haidian District, 100872 Beijing, China. Hauptthemen laut MEDLINE: Proteomics, Neural Networks, Computer, Proteins, Computational Biology. Die Arbeit nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Key Research and Development Program, National Natural Science Foundation of China, Shenzhen Medical Research Fund.
Abstract (Originalsprache)
Advances in high-resolution proteomic technologies are enabling systematic characterization of cellular heterogeneity and protein functional programs at increasing depth. However, existing computational approaches typically focus on either cell-level organization or protein-level structure, lacking a unified framework that jointly models both entities. Here we present scYOU, a multi-scale yoked optimization and unsupervised framework that simultaneously captures cellular heterogeneity and protein functional organization through a dual-channel neural network architecture. scYOU learns cell and protein embeddings in parallel while coupling the two views through a supercell-constrained alignment strategy. Supercells are constructed based on high-order expression similarity and define grouping structures over cells. A dual-channel alignment objective is imposed at the reconstruction level, such that reconstructed expression profiles of cells within the same supercell are encouraged to converge, whereas those from different supercells remain separable. On the protein side, Gene Ontology–derived functional similarity priors regularize protein embeddings to preserve molecular biological consistency. Across diverse benchmark datasets, scYOU achieves strong and relatively consistent clustering performance together with improved embedding separability, while the learned protein representations support embedding-guided refinement of existing interaction networks and the recovery of complex-associated protein modules. These results establish scYOU as a structurally coupled, multi-level representation learning framework for proteomic data, and highlight the importance of jointly modeling cellular heterogeneity and protein functional modules within a unified computational paradigm. Our code and data are available at https://github.com/anchor-hue/scYou.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Substanzen
Autor:innen und Institutionen
- School of Mathematics, Renmin University of China, No. 59 Zhongguancun Street, Haidian District, 100872 Beijing, China.
Förderung
- National Key Research and Development Program
- National Natural Science Foundation of China
- Shenzhen Medical Research Fund
- Renmin University of China
Zitieren
APA
Gan, Y., Huang, Y., & Gong, X. (2026). scYOU: a multi-scale yoked optimization and unsupervised framework for analysis of cellular heterogeneity and complex-associated protein module recovery. Briefings in Bioinformatics, 27(6). https://doi.org/10.1093/bib/bbag626
Vancouver
Gan Y, Huang Y, Gong X. scYOU: a multi-scale yoked optimization and unsupervised framework for analysis of cellular heterogeneity and complex-associated protein module recovery. Briefings in Bioinformatics. 2026;27(6). doi:10.1093/bib/bbag626
BibTeX
@article{Gan2026,
author = {Gan, Yuan and Huang, Yixiang and Gong, Xinqi},
title = {scYOU: a multi-scale yoked optimization and unsupervised framework for analysis of cellular heterogeneity and complex-associated protein module recovery},
journal = {Briefings in Bioinformatics},
year = {2026},
volume = {27},
number = {6},
doi = {10.1093/bib/bbag626}
}
EUVIMED