CanCellCap: robust cancer cell capture across tissue types on single-cell RNA-seq data by multi-domain learning

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cancer cellular diversity but faces challenges in accurate cancer cell identification due to heterogeneity.
• Gene expression variability hinders the generalization and robustness of existing methods for cancer cell detection.

Purpose of the Study:

• To develop a universal framework, CanCellCap, for identifying cancer cells in scRNA-seq data across diverse tissues, cancers, and sequencing platforms.
• To enhance the robustness and generalization capabilities of cancer cell identification methods.

Main Methods:

• CanCellCap employs a multi-domain learning framework integrating domain adversarial learning and Mixture of Experts.
• A masking-reconstruction strategy is utilized to handle data from different sequencing platforms.
• The framework extracts common and specific gene expression patterns for cancer and normal cells across tissues.

Main Results:

• CanCellCap achieved an average accuracy of 0.977 in cancer cell identification across 13 tissue types, 23 cancer types, and 7 sequencing platforms.
• Outperformed five state-of-the-art methods on 33 benchmark datasets, demonstrating superior performance.
• Showcased high performance on unseen cancer types, tissues, and species, and accurately identified cancer spots in spatial transcriptomics data.
• Demonstrated computational efficiency, analyzing 100,000 cells in minutes, and revealed critical biomarkers and pathways.

Conclusions:

• CanCellCap offers a robust and accurate solution for cancer cell identification across diverse scRNA-seq data.
• Its strong generalization to unseen data and adaptability to spatial transcriptomics highlight its versatility for research and clinical applications.