Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Spatial Transcriptomics as Images for Large-Scale Pretraining

Yishun Zhu, Jiaxin Qi, Jian Wang, Yuhua Zheng, Jianqiang Huang

Abstract

Spatial Transcriptomics (ST) profiles thousands of gene expression values at discrete spots with precise coordinates on tissue sections, preserving spatial context essential for clinical and pathological studies. With rising sequencing throughput and advancing platforms, the expanding data volumes motivate large-scale ST pretraining. However, the fundamental unit for pretraining, i.e., what constitutes a single training sample, remains ill-posed. Existing choices fall into two camps: (1) treating each spot as an independent sample, which discards spatial dependencies and collapses ST into single-cell transcriptomics; and (2) treating an entire slide as a single sample, which produces prohibitively large inputs and drastically fewer training examples, undermining effective pretraining. To address this gap, we propose treating spatial transcriptomics as croppable images. Specifically, we define a multi-channel image representation with fixed spatial size by cropping patches from raw slides, thereby preserving spatial context while substantially increasing the number of training samples. Along the channel dimension, we define gene subset selection rules to control input dimensionality and improve pretraining stability. Extensive experiments show that the proposed image-like dataset construction for ST pretraining consistently improves downstream performance, outperforming conventional pretraining schemes. Ablation studies verify that both spatial patching and channel design are necessary, establishing a unified, practical paradigm for organizing ST data and enabling large-scale pretraining.

Spatial Transcriptomics as Images for Large-Scale Pretraining

Abstract

Spatial Transcriptomics (ST) profiles thousands of gene expression values at discrete spots with precise coordinates on tissue sections, preserving spatial context essential for clinical and pathological studies. With rising sequencing throughput and advancing platforms, the expanding data volumes motivate large-scale ST pretraining. However, the fundamental unit for pretraining, i.e., what constitutes a single training sample, remains ill-posed. Existing choices fall into two camps: (1) treating each spot as an independent sample, which discards spatial dependencies and collapses ST into single-cell transcriptomics; and (2) treating an entire slide as a single sample, which produces prohibitively large inputs and drastically fewer training examples, undermining effective pretraining. To address this gap, we propose treating spatial transcriptomics as croppable images. Specifically, we define a multi-channel image representation with fixed spatial size by cropping patches from raw slides, thereby preserving spatial context while substantially increasing the number of training samples. Along the channel dimension, we define gene subset selection rules to control input dimensionality and improve pretraining stability. Extensive experiments show that the proposed image-like dataset construction for ST pretraining consistently improves downstream performance, outperforming conventional pretraining schemes. Ablation studies verify that both spatial patching and channel design are necessary, establishing a unified, practical paradigm for organizing ST data and enabling large-scale pretraining.
Paper Structure (10 sections, 8 equations, 3 figures, 9 tables, 1 algorithm)

This paper contains 10 sections, 8 equations, 3 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Method
  4. Preliminaries
  5. Our Method
  6. Experiments
  7. Datasets
  8. Implementation Details
  9. Results and Analysis
  10. Conclusion

Figures (3)

  • Figure 1: Illustration of data organization strategies for large-scale spatial transcriptomics pretraining. (a) Spot-based method: each spot is treated as an independent $K$-dimensional gene-expression vector, discarding spatial relationships. (b) Slice-based method: each whole tissue slice is treated as a single training sample, leading to very large inputs and few examples. (c) Ours: spatial transcriptomics is treated as croppable multi-channel images by extracting fixed-size spatial patches ($h \times w$) with selected gene subsets along the channel dimension ($k$), preserving local spatial context while greatly increasing the number of training samples.
  • Figure 2: Sample construction — ori (original), compacted (scaled), cropped (fixed-window), normalized (grid-mapped). Visualization: spot size 100 for ori, 1 for others.
  • Figure 3: Visualization of BrC results (Tables \ref{['tab1']} and Tables \ref{['tab:VSsToFM']}) — red boxes indicate largest improvements.