Page 34 - TD-3-1

Tumor Discovery AI uncovers tumor spatial organization

A

B C

D

Figure 4. The proposed spatial clustering method, VGAE_SGC, undergoes validation using single-cell resolution breast cancer spatial transcriptomics (ST)
data: (A) The primary breast cancer dataset is obtained through the 10× Xenium technology. To enhance computational efficiency, a cropped subset of this
ST data is utilized in this study; (B) the adjusted Rand index (ARI) values for seven compared approaches are computed using the cropped ST data; (C)
the spatial distribution of spots in the VGAE_SGC method yields an ARI value of 0.409; and (D) the reduced-dimension graph illustrates VGAE_SGC’s
spatial clustering outcomes and its associated inferred spatial evolutionary trajectory.

extensive single-cell data, which is a significant factor in clustering methodologies. We validated our proposed
optimizing the performance of our algorithm. method, VGAE_SGC, using breast cancer datasets with
varying resolutions. In addition, we delineated various
4. Conclusion downstream analysis tasks predicted for spatial clustering
This article explores the utilization of artificial outcomes, including differential gene expression analysis,
intelligence algorithms for spatial clustering tasks in identification of biological functions, and inference
spatially resolved transcriptomics. Specifically, we of spatial developmental trajectories. Unsupervised
introduced a VGAE enhanced with SGC ConvNets to learning modules face challenges in accurately identifying
delineate spatial domains in ST data from various tumor spatial regions within an organization. However, future
tissues. The clustering accuracy of our architecture is research can focus on integrating methods to enhance
demonstrated through comparisons with several existing the clustering results. Approaches such as contrastive and

Volume 3 Issue 1 (2024) 8 https://doi.org/10.36922/td.2049