scPanel: a tool for automatic identification of sparse gene panels for generalizable patient classification using...

Area of Science:

• Computational biology
• Genomics
• Biomarker discovery

Background:

• Single-cell RNA sequencing (scRNA-seq) enables transcriptomic profiling at single-cell resolution in large patient cohorts.
• Discovery of gene and cellular biomarkers is crucial for disease understanding and patient stratification.
• High costs associated with large gene panels hinder clinical translation of scRNA-seq findings.

Purpose of the Study:

• To introduce scPanel, a computational framework for identifying sparse gene panels for patient classification.
• To bridge the gap between biomarker discovery and clinical application by reducing gene panel size.
• To enable accurate patient classification using minimal, automatically selected informative biomarker genes.

Main Methods:

• scPanel identifies informative biomarker genes from cell populations most responsive to perturbations.
• A data-driven approach automatically determines a minimal set of biomarker genes.
• Patient-level classification is achieved by aggregating cell-associated prediction probabilities using the area under the curve score.

Main Results:

• scPanel achieved high patient classification accuracy in scleroderma, colorectal cancer, and COVID-19 datasets using <20 genes.
• The framework demonstrated cross-dataset generalizability for predicting disease state in an external COVID-19 cohort.
• scPanel outperformed existing gene selection methods for patient classification.

Conclusions:

• scPanel effectively identifies parsimonious sets of reliable biomarker candidates for clinical translation.
• The framework reduces the cost and complexity of translating scRNA-seq discoveries into clinical applications.
• scPanel facilitates the development of cost-effective diagnostic and prognostic tools based on transcriptomic data.