Deep learning using bulk RNA-seq data expands cell landscape identification in tumor microenvironment.

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• The tumor microenvironment (TME) is critical for tumor progression and immunotherapy outcomes.
• Current methods for analyzing TME heterogeneity are limited in scope and cell-type resolution.
• Understanding TME complexity is essential for advancing cancer immunotherapy.

Purpose of the Study:

• To develop a novel deep learning framework, DCNet, for comprehensive TME cell landscape inference.
• To enable the identification of over 400 cell types from bulk RNA-sequencing (RNA-seq) data.
• To provide a robust tool for decoding tumor microenvironment heterogeneity.

Main Methods:

• Developed DCNet, a deep learning model embedding cell-gene relationships.
• Utilized bulk RNA-seq data to infer detailed cell landscapes.
• Validated DCNet against single-cell RNA-seq datasets for accuracy and stability.

Main Results:

• DCNet accurately recapitulated cell landscapes from multiple single-cell RNA-seq datasets.
• The model demonstrated robustness and stability in TME heterogeneity analysis.
• Analysis of TCGA patient data revealed distinct patient groups with significant survival differences based on inferred cell populations.

Conclusions:

• DCNet offers a powerful foundation for dissecting tumor microenvironment heterogeneity.
• The tool enhances understanding of TME complexity and its impact on patient survival.
• DCNet facilitates the development of more effective immunotherapies by providing a detailed cell landscape.