CaSee

CaSee discriminates cancer cells from normal cells in single-cell RNA sequencing (scRNA-seq) expression matrices using transfer learning from pan-cancer bulk sequencing data.

Key Features:

• Transfer Learning Approach: Applies transfer learning from pan-cancer bulk sequencing data to scRNA-seq to improve discrimination between cancer and normal cells.
• Comparison to CNV Analysis: Delivers discrimination performance beyond conventional copy number variation (CNV) analysis.
• Supported scRNA-seq Data Types: Supports scRNA tissue sequencing data, scRNA cell line sequencing data, scRNA xenograft cell sequencing data, and scRNA circulating tumor cell sequencing data.
• Compatible Sequencing Platforms: Compatible with 10× Genomics Chromium, Smart-seq2, and Microwell-seq.

Scientific Applications:

• Intratumoral Heterogeneity: Enables identification of malignant versus non-malignant cells to characterize cellular diversity within individual tumors.
• Intertumoral Heterogeneity: Facilitates comparison of cancer cell signatures across tumors and cohorts to study between-tumor variability.
• Cellular Composition for Targeted Therapies: Supports delineation of tumor cellular composition to inform studies related to targeted therapeutic strategies.

Methodology:

Deep-learning-based model that leverages transfer learning from pan-cancer bulk sequencing data applied to scRNA-seq expression matrices; evaluated in multicenter assessments across 11 retrospective cohorts and an independent dataset, achieving an average discrimination accuracy of 96.69%.