Page 27 - TD-3-1

Tumor Discovery

ORIGINAL RESEARCH ARTICLE
Artificial intelligence enabled spatially resolved

transcriptomics reveal spatial tissue organization
of multiple tumors

2
1†
1
Teng Liu 1,2† , Jinxin Ye , Chunnan Hu , Zongbo Zhang , Zhuomiao Ye ,
3†
1,2
2
Jiangnan Liao , and Mingzhu Yin *
1 Department of Clinical Research Center (CRC), Clinical Pathology Center (CPC), Cancer Early
Detection and Treatment Center (CEDTC) and Translational Medicine Research Center (TMRC),
Chongqing University Three Gorges Hospital, Chongqing University, Wanzhou, Chongqing, China
2 Department of Chongqing Technical Innovation Center for Quality Evaluation and Identification of
Authentic Medicinal Herbs, Chongqing University Three Gorges Hospital, Chongqing University,
Wanzhou, Chongqing, China
3 Physics and Technique Department of Radiation Oncology, Cancer treatment institute,
Chongqing University Three Gorges Hospital, Chongqing University, Wanzhou, Chongqing, China

Abstract

Spatially resolved transcriptomics was honored as the Method of the Year 2020
by Nature Methods. This approach allows biologists to precisely discern mRNA
expression at the cellular level within structurally preserved tissues. Leveraging
artificial intelligence in spatial transcriptomic analysis enhances the understanding
of cellular-level biological interactions and offers novel insights into intricate tissues,
† These authors contributed equally
to this work. such as tumor microenvironments. Nevertheless, numerous existing clustering
algorithms employing deep learning exhibit the potential for enhancement. In this
*Corresponding author: paper, we focus on graph deep learning-based spatial domain identification for
Mingzhu Yin
(yinminzhu@cqu.edu.cn) spatial transcriptomics (ST) data from multiple tumors. This identification enables
the recognition of cell subpopulations in distinct spatial coordinates, aiding further
Citation: Liu T, Ye J, Hu C,
et al. Artificial intelligence enabled studies on tumor progression, such as cell-cell communication, pseudo-time
spatially resolved transcriptomics trajectory inference, and single-cell deconvolution. Initially, the gene expression
reveal spatial tissue organization profiles and spatial location information were transformed into a gene feature matrix
of multiple tumors. Tumor Discov.
2024;3(1):2049. and a cell adjacency matrix. A variational graph autoencoder was then applied
https://doi.org/10.36922/td.2049 to extract features and reduce the dimensions of these two matrices. Following
Received: October 16, 2023 training in the constructed graph neural networks, the latent embeddings of ST data
Accepted: December 15, 2023 were generated and could be leveraged for spatial domain identification. Through
Published Online: March 6, 2024 a comparison with established methods, our approach demonstrated superior
Copyright: © 2024 Author(s). clustering accuracy. The utilization of accurately segmented spatial regions enables
This is an Open-Access article downstream analyses of multiple tumors, encompassing the trajectory of tumor
distributed under the terms of the
Creative Commons Attribution evolution, and facilitating differential gene expression analysis across various cell
License, permitting distribution, types.
and reproduction in any medium,
provided the original work is
properly cited. Keywords: Spatial transcriptomics; Artificial intelligence; Graph neural network; Spatial
Publisher’s Note: AccScience domain identification; Tumor progression
Publishing remains neutral with
regard to jurisdictional claims in
published maps and institutional
affiliations.

Volume 3 Issue 1 (2024) 1 https://doi.org/10.36922/td.2049