scAURA: Alignment- and Uniformity-based Graph Debiased Contrastive Representation Architecture for Self-Supervised...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution transcriptomic data for understanding cellular heterogeneity.
• Accurate cell type identification from scRNA-seq data is hindered by data challenges like high dimensionality, sparsity, and noise.

Purpose of the Study:

• To develop a robust and accurate computational framework for cell type identification in scRNA-seq data.
• To address the limitations of existing methods in handling noisy and high-dimensional scRNA-seq datasets.

Main Methods:

• Introduction of scAURA (single cell Alignment- and Uniformity-based Graph Debiased Contrastive Representation Architecture).
• Integration of graph debiased contrastive learning with self-supervised clustering for unified cell type identification.
• Evaluation on 18 diverse scRNA-seq datasets across multiple platforms and species (human and mouse).

Main Results:

• scAURA demonstrated superior performance compared to state-of-the-art methods, achieving top ranks in Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI) across multiple datasets.
• The framework exhibited strong robustness against dropout noise and sparsity, maintaining stable clustering performance.
• Application to an Alzheimer's disease dataset successfully clustered cell types, identified novel marker genes, and inferred transcriptional regulators.

Conclusions:

• scAURA provides a consistent and superior approach for cell type identification in scRNA-seq data.
• The method's robustness makes it suitable for analyzing challenging and noisy single-cell datasets.
• scAURA has potential applications in disease research, including identifying cell-specific mechanisms in neurodegenerative disorders.