Graph embedding and Gaussian mixture variational autoencoder network for end-to-end analysis of single-cell RNA...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high resolution for cellular gene expression but suffers from technical limitations causing sparse and heterogeneous data.
• Identifying distinct cell subpopulations and understanding complex biological processes requires robust methods to handle scRNA-seq data challenges.

Purpose of the Study:

• To introduce autoCell, a novel deep learning framework for scRNA-seq data.
• To enhance scRNA-seq data quality through dropout imputation and feature extraction.
• To enable comprehensive analysis, including visualization, clustering, and disease-specific network identification.

Main Methods:

• Developed autoCell, a variational autoencoding network integrating graph embedding and a probabilistic depth Gaussian mixture model.
• Inferred the distribution of high-dimensional, sparse scRNA-seq data.
• Validated autoCell on simulated and real-world scRNA-seq datasets.

Main Results:

• autoCell effectively imputes missing gene expression data, improving downstream analyses.
• The tool enhanced the identification of cell developmental trajectories in human preimplantation embryos.
• Identified disease-associated astrocytes (DAAs) and reconstructed their specific molecular networks and cell-cell communication pathways in Alzheimer's disease.

Conclusions:

• autoCell provides a powerful toolbox for end-to-end scRNA-seq data analysis.
• The imputation and feature extraction capabilities of autoCell facilitate deeper biological insights, particularly in disease contexts.
• This approach advances the utility of scRNA-seq for understanding cellular heterogeneity and disease mechanisms.