scBGEDA

scBGEDA performs clustering and marker-gene analysis of single-cell RNA sequencing (scRNA-seq) data by combining a dual denoising autoencoder with bipartite graph ensemble clustering to identify cell types and quantify gene influence.

Key Features:

• Dual Denoising Autoencoder Network: Employs a dual denoising autoencoder to project high-dimensional scRNA-seq data into a compressed latent space and optimizes feature representation using zero-inflated negative binomial reconstruction loss and denoising reconstruction loss.
• Bipartite Graph Ensemble Clustering: Uses a bipartite graph ensemble clustering algorithm with a graph-based consensus function to exploit relationships between cells and latent representations for robust clustering.
• Performance and Scalability: Demonstrated performance across 20 diverse scRNA-seq datasets from various sequencing platforms and scalable behavior on large-scale datasets relative to state-of-the-art methods.
• Biological Insights: Identifies cell-type-specific marker genes and quantifies gene influence on clusters to support downstream functional genomic analyses.

Scientific Applications:

• Cell Type Identification: Accurate clustering of heterogeneous scRNA-seq samples to delineate distinct cell types.
• Transcriptome Characterization: Identification of marker genes and quantification of gene influence to characterize cellular transcriptomes.
• Functional Genomics: Support for functional genomic studies exploring gene roles and interactions across cell populations.

Methodology:

Data projection via a dual denoising autoencoder using zero-inflated negative binomial and denoising reconstruction losses; bipartite graph ensemble clustering with a graph-based consensus function to derive consensus partitions from latent representations; and empirical validation across multiple scRNA-seq datasets.