henhaobio
Sign in

bioRxivBioinformaticsPreprint

Unsupervised Spatial Embedded Deep Representation of Spatial Transcriptomics

Huazhu FuiD, Hang XuiD, Kelvin Chong, Mengwei LiiD, Kok Siong AngiD, Hong Kai LeeiD, Jingjing Ling, Ao CheniD, Ling ShaoiD, Longqi LiuiD, Jinmiao CheniD

PDFDOI
Structured summary available · Jump to summary ↓

Abstract

Spatial transcriptomics enable us to dissect tissue heterogeneity and map out inter-cellular communications. Optimal integration of transcriptomics data and associated spatial information is essential towards fully exploiting the data. We present SEDR, an unsupervised spatially embedded deep representation of both transcript and spatial information. The SEDR pipeline uses a deep autoencoder to construct a low-dimensional latent representation of gene expression, which is then simultaneously embedded with the corresponding spatial information through a variational graph autoencoder. We applied SEDR on human dorsolateral prefrontal cortex data and achieved better clustering accuracy, and correctly retraced the prenatal cortex development order with trajectory analysis. We also found the SEDR representation to be eminently suited for batch integration. Applying SEDR to human breast cancer data, we discerned heterogeneous sub-regions within a visually homogenous tumor region, identifying a tumor core with pro-inflammatory microenvironment and an outer ring region enriched with tumor associated macrophages which drives an immune suppressive microenvironment.

Abstract summary

Claude Opus 5.5 · xhigh effort · 18 Oct 2026 · from the abstract · CC BY-NC-ND 4.0

SEDR couples a deep autoencoder over expression with a variational graph autoencoder over position, improving clustering in dorsolateral prefrontal cortex and splitting a visually uniform breast tumour region.

Spatial transcriptomics can dissect tissue heterogeneity and map inter-cellular communication, and integrating expression with the associated spatial information is called essential to exploiting the data fully. SEDR learns one unsupervised embedding that joins a latent expression code with the spatial information through a variational graph autoencoder, then reuses it for clustering, trajectories and batch integration.

Key results

  1. SuggestedOn human dorsolateral prefrontal cortex data the embedding gave more accurate clustering than the comparison, and a trajectory analysis recovered the expected ordering of prenatal cortical development.

    Reported as better clustering accuracy and a correctly retraced developmental order; the abstract names no accuracy metric, no value and no competing method. · Abstract

  2. SuggestedIn human breast cancer tissue the representation broke a region that looks uniform into sub-regions: a core whose microenvironment is pro-inflammatory, and a surrounding ring rich in tumour-associated macrophages linked to immune suppression.

    Described as a single case study in the abstract, with no sub-region sizes, marker statistics or independent validation reported. · Abstract

  3. ClaimedThe same representation is put forward as a good basis for integrating batches.

    Asserted in the abstract without naming datasets, integration metrics or methods compared against. · Abstract

Methods, data and reproducibility

A deep autoencoder compresses gene expression into a low-dimensional latent code, which a variational graph autoencoder then embeds jointly with the corresponding spatial information, yielding one unsupervised representation used downstream for clustering, trajectory analysis and batch integration.

  • Spatial transcriptomics of dorsolateral prefrontal cortex, used for clustering and trajectory analysis · human
  • Spatial transcriptomics of breast cancer tissue, used for the sub-region analysis · human
Methods & tools
SEDR, deep autoencoder, variational graph autoencoder
Reproducibility gaps
No repository address, dataset accession, framework or version information appears in the abstract.
Limitations (2)
  • Clustering is said to be more accurate, but no metric, value or named baseline is given, so the size of the gain cannot be judged. · noted in this summary
  • The tumour sub-regions are given a biological reading with no orthogonal validation, such as staining or an independent cohort, reported in the abstract. · noted in this summary

Read this if you cluster spatial transcriptomics data and want a joint expression-and-graph embedding that is also put forward for trajectories and batch integration.

Citations
102

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISYu '24Chang '22Li '25Shang '22Long '23Dong '22
65% semantic matchBioinformatics
Yuanyuan Yu, Yao He, Zhi Xie·Biomolecules·Jun 9, 2024· 6 citations