CanSig Benchmarks Methods for Reproducible Cancer Cell State Discovery from Single-Cell Transcriptomic Data.

Area of Science:

• Computational biology
• Cancer research
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals gene expression signatures for cell states, crucial for precision oncology.
• Lack of standardized computational methods hinders reproducibility in scRNA-seq signature detection.

Purpose of the Study:

• To develop CanSig, a comprehensive benchmarking tool for evaluating methods identifying transcriptional signatures in cancer.
• To provide a standardized and reproducible framework for uncovering clinically relevant cancer cell states from scRNA-seq data.

Main Methods:

• CanSig integrates metrics for batch correction, biological signal conservation, and signature correlation.
• Evaluated thirteen computational methods on twelve scRNA-seq datasets across five human cancer types (glioblastoma, breast, lung, rhabdomyosarcoma, squamous cell carcinoma).
• Analyzed data from 185 patients, encompassing 174,000 malignant cells.

Main Results:

• Identified gene expression signatures correlating with clinical outcomes like patient survival and lymph node metastasis.
• Harmony, BBKNN, and fastMNN emerged as top-performing integration methods for discovering shared cancer cell states.
• Demonstrated the clinical relevance of scRNA-seq-derived signatures.

Conclusions:

• CanSig offers a standardized and reproducible approach for analyzing cancer scRNA-seq data.
• Facilitates the discovery of clinically relevant cancer cell states, advancing patient stratification and precision oncology.
• Highlights the importance of robust computational methods for reliable signature detection.