scSurvival: Single-Cell Survival Analysis of Clinical Cancer Cohort Data at Cellular Resolution

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Survival analysis is crucial in cancer research.
• Single-cell sequencing (scRNA-seq) is increasingly used with clinical data.
• Existing methods lack direct survival outcome modeling from single-cell data.

Purpose of the Study:

• To develop an effective strategy for modeling survival outcomes directly from single-cell data.
• To introduce scSurvival, an attention-based multiple-instance Cox regression framework.
• To enable robust prediction of patient survival and identification of survival-associated cell subpopulations.

Main Methods:

• scSurvival utilizes an attention-based multiple-instance Cox regression framework.
• It models each tumor sample as an ensemble of cells.
• Integrates variational autoencoder (VAE) for feature extraction to handle high dimensionality, sparsity, and batch effects.

Main Results:

• scSurvival demonstrates superior performance and scalability in simulations.
• Accurately predicts patient outcomes in melanoma and liver cancer scRNA-seq cohorts.
• Identifies cell subpopulations critical to patient survival.

Conclusions:

• scSurvival enables robust prediction of patient survival from single-cell data.
• Uncovers survival-associated cell subpopulations, advancing single-cell survival analysis.
• Promotes broader adoption of cohort-level single-cell profiling in cancer research.