Skip to content
DraconiticResearch
Search Sources API
194,226 studies31 sourcesUpdated 4 min ago
Peer-reviewedCC BY 4.0Nature Communications

DR-GEM enables self-supervised machine learning for single-cell embeddings and annotations

  • Christine Yiwen Yeh1,2,3
  • Min Woo Sun1
  • Dixian Zhu2
  • Livnat Jerby2,4,5
5 affiliations
  1. 1Department of Biomedical Data Science, Stanford University School of Medicine, Stanford, CA USA
  2. 2Department of Genetics, Stanford University School of Medicine, Stanford, CA USA
  3. 3Department of Medicine, Stanford University School of Medicine, Stanford, CA USA
  4. 4Stanford Cancer Institute, Stanford University School of Medicine, Stanford, CA USA
  5. 5Chan Zuckerberg Biohub, San Francisco, CA USA
Read full textPDFDOICite

Abstract

Dimensionality reduction and clustering are instrumental to single-cell and spatial genomics data analysis. Here we show that existing methods tend to fit oversampled classes and miss more unique signals and patterns, such as rare cell types and states. We find that the problem in such cases is not primarily data scarcity per se, but the model equating abundance with importance. Addressing this, we present “Distributionally Robust and latent Group-AwarE consensus Machine learning” or DR-GEM, a self-supervised meta-algorithm that uses the reconstruction error to reorient the model attention to patterns it initially missed and applies balanced consensus learning to increase robustness and filter low-quality data. We apply DR-GEM to synthetic and real-world single-cell, spatial transcriptomics, and Perturb-seq datasets and demonstrate that it outperforms existing methods in obtaining reliable embeddings, recovering rare cell types, filtering noise, mitigating cell-level and patient-level underrepresentation, and uncovering held-out genetic and spatial features. We thus surface and address an underappreciated problem and present a self-supervised framework to help support AI-based and data-driven discoveries.

Related

  1. Representation learning for low-data protein fitness prediction using sequence–structure embeddings and graph neural networks
    Chanoknun Phunnasorn, Narupat Hongdilokkul, Wasakorn Laesanklang +2
    Scientific ReportsPeer-reviewedCC BY-NC-ND 4.0Full textMolecular biology & gene regulationCrossref
  2. METANet: a supervised ensemble learning framework for reconstructing direct and functional tissue-specific transcription factor networks.
    Wooseok J Jung, Sandeep Acharya, Daniel P Ruskin +4
    Bioinformatics (Oxford, England)Peer-reviewedCC BYFull text2 citedMolecular biology & gene regulationEurope PMCCrossref
  3. DNA augmentations for semi-supervised learning in genomic sequence classification.
    Hyunwoo Yoo, Gail L Rosen
    NAR genomics and bioinformaticsPeer-reviewedCC BY-NCFull textMolecular biology & gene regulationEurope PMCCrossref
  4. Deciphering ribosomal frameshifting determinants across species with a semi-supervised hybrid learning framework
    Yu-Hao Yang, Juan Yang, Zi-Jia Liu +4
    Zoological ResearchPeer-reviewedCC BY-NC 4.0Full textMolecular biology & gene regulationPMC
  5. EnzyREx: deep learning enables reaction-based comprehensive enzyme screening
    Keisuke Hirota, Takuji Yamada
    Genome BiologyPeer-reviewedCC BY 4.0Molecular biology & gene regulationGenetics & genomicsCrossref

Cite

BibTeXRIS
BibTeX
@article{yeh2026drgem,
  title = {{DR-GEM enables self-supervised machine learning for single-cell embeddings and annotations}},
  author = {Christine Yiwen Yeh and Min Woo Sun and Dixian Zhu and Livnat Jerby},
  journal = {Nature Communications},
  year = {2026},
  month = sep,
  doi = {10.1038/s41467-026-77908-z},
  url = {https://doi.org/10.1038/s41467-026-77908-z},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - DR-GEM enables self-supervised machine learning for single-cell embeddings and annotations
AU  - Christine Yiwen Yeh
AU  - Min Woo Sun
AU  - Dixian Zhu
AU  - Livnat Jerby
PY  - 2026
DA  - 2026/09/24
JO  - Nature Communications
DO  - 10.1038/s41467-026-77908-z
UR  - https://doi.org/10.1038/s41467-026-77908-z
AB  - Dimensionality reduction and clustering are instrumental to single-cell and spatial genomics data analysis. Here we show that existing methods tend to fit oversampled classes and miss more unique signals and patterns, such as rare cell types and states. We find that the problem in such cases is not primarily data scarcity per se, but the model equating abundance with importance. Addressing this, we present “Distributionally Robust and latent Group-AwarE consensus Machine learning” or DR-GEM, a self-supervised meta-algorithm that uses the reconstruction error to reorient the model attention to patterns it initially missed and applies balanced consensus learning to increase robustness and filter low-quality data. We apply DR-GEM to synthetic and real-world single-cell, spatial transcriptomics, and Perturb-seq datasets and demonstrate that it outperforms existing methods in obtaining reliable embeddings, recovering rare cell types, filtering noise, mitigating cell-level and patient-level underrepresentation, and uncovering held-out genetic and spatial features. We thus surface and address an underappreciated problem and present a self-supervised framework to help support AI-based and data-driven discoveries.
SN  - 2041-1723
ER  -