GTADC: A Graph-Based Method for Inferring Cell Spatial Distribution in Cancer Tissues.

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Tumor heterogeneity complicates the analysis of cellular interactions and ecosystem construction in cancer tissues.
• Integrating spatial transcriptomics (ST) and single-cell sequencing (scRNA-seq) is crucial for detailed analysis.
• Traditional deep learning methods struggle with gene selection in heterogeneous cancer tissues, limiting cell type differentiation.

Purpose of the Study:

• To develop a novel graph-based deep learning method, GTADC, for precise cell type identification and spatial composition analysis in tumors.
• To overcome limitations of traditional methods in capturing gene expression differences within heterogeneous cancer cell populations.
• To enhance the accuracy of gene selection using Silhouette scores for improved resolution of cell types.

Main Methods:

• Proposed GTADC, a graph-based deep learning approach utilizing Silhouette scores for enhanced gene selection.
• Incorporated graph structures to capture spatial and topological relationships between ST and scRNA-seq data.
• Evaluated GTADC's ability to consider intra-cluster gene similarity and overall clustering structure.

Main Results:

• GTADC precisely captures genes with significant expression differences within cell types, improving accuracy.
• The method effectively resolves the spatial composition of different cell types within tissues by leveraging graph structures.
• GTADC enables accurate reconstruction of cellular spatial composition and identification of potential cancer cell regions.

Conclusions:

• GTADC offers a precise solution for inferring cell spatial composition and understanding tumor ecosystems.
• The method facilitates early detection of cancer regions by assessing quantity and spatial information.
• This approach contributes to a deeper understanding of early-stage cancer and supports early diagnosis efforts.