Supervised deep learning with gene functional annotation for cell classification.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA-sequencing (scRNA-seq) analysis commonly uses gene-by-gene differential expression.
• Large scRNA-seq datasets yield many significant but biologically irrelevant genes, hindering interpretation.
• Existing methods struggle to balance accurate classification with biologically meaningful gene set identification.

Purpose of the Study:

• To develop a novel method, Supervised Deep learning with gene functional ANnotation (SDAN), for scRNA-seq data interpretation.
• To integrate gene functional annotation with gene-expression profiles for improved biological insights.
• To identify functionally coherent gene sets that accurately classify cells and predict outcomes.

Main Methods:

• Developed SDAN, a graph neural network-based approach.
• Integrated gene functional annotation (e.g., protein-protein interactions) with gene-expression data.
• Applied SDAN to scRNA-seq datasets for severe COVID-19, dementia, and cancer immunotherapy response.

Main Results:

• SDAN successfully identified functionally coherent gene sets across diverse applications.
• The method achieved accurate outcome classification simultaneously with interpretable gene assignments.
• SDAN outperformed three representative existing methods in all evaluated real-data applications.

Conclusions:

• SDAN offers a powerful and interpretable approach for analyzing complex scRNA-seq data.
• The integration of functional genomics enhances the biological relevance of findings from scRNA-seq studies.
• SDAN provides a robust framework for predicting disease severity and treatment response using scRNA-seq.