TUSCAN: Tumor segmentation and classification analysis in spatial transcriptomics

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Accurate identification of tumor cells is crucial for understanding tumor heterogeneity and the tumor microenvironment.
• Spatially resolved transcriptomics (SRT) enables transcript quantification in intact tissues but faces challenges in precise tumor region detection.
• Current methods rely on marker genes or copy number alterations, with limitations in accuracy and applicability.

Purpose of the Study:

• To introduce TUSCAN (TUmor Segmentation and Classification ANalysis), a novel computational method for accurate tumor region identification in spatial transcriptomics data.
• To improve the delineation of tumor sections and benign tissues by integrating SRT gene expression and histology imaging.
• To provide interpretable clonal evolution inferences for novel insights into cancer development and potential therapeutic targets.

Main Methods:

• TUSCAN constructs a spatial copy number variation profile using SRT data.
• It integrates gene information from SRT with hematoxylin-and-eosin staining images for annotation.
• Performance was benchmarked against existing methods across multiple SRT datasets and platforms.

Main Results:

• TUSCAN effectively delineates tumor regions with improved accuracy compared to existing approaches.
• The method demonstrates robust performance across diverse datasets and SRT platforms.
• Output provides interpretable insights into clonal evolution within tumor tissues.

Conclusions:

• TUSCAN offers a significant advancement in accurately segmenting tumor regions from spatial transcriptomics data.
• The integration of copy number variation profiles enhances tumor identification accuracy.
• The tool facilitates deeper understanding of tumor biology, clonal heterogeneity, and potential therapeutic strategies.