DeepGSEA: explainable deep gene set enrichment analysis for single-cell transcriptomic data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Gene set enrichment (GSE) analysis is crucial for interpreting gene expression and understanding biological phenotypes.
• Single-cell RNA sequencing (scRNA-seq) enables fine-grained GSE analysis but faces challenges due to cellular heterogeneity.
• Current statistical GSE methods can struggle to identify enriched gene sets in complex scRNA-seq data.

Purpose of the Study:

• To develop an explainable deep learning approach for gene set enrichment analysis in scRNA-seq data.
• To address the interpretability challenges of deep learning in GSE analysis.
• To enhance the identification of enriched gene sets in heterogeneous single-cell data.

Main Methods:

• Introduced DeepGSEA, an explainable deep learning framework utilizing interpretable, prototype-based neural networks.
• Designed classification tasks for DeepGSEA to learn and capture gene set enrichment information.
• Enabled significance testing on individual gene sets and visualization of gene set distributions via embeddings.

Main Results:

• DeepGSEA demonstrates superior sensitivity and specificity compared to conventional GSE methods in simulation studies.
• The approach was validated on three real-world scRNA-seq datasets.
• The interpretability of DeepGSEA was illustrated by explaining its analysis results.

Conclusions:

• DeepGSEA offers a powerful and interpretable solution for gene set enrichment analysis in scRNA-seq data.
• The method enhances the identification of biologically relevant gene sets amidst cellular heterogeneity.
• DeepGSEA provides a new avenue for leveraging deep learning in single-cell omics data interpretation.