scDeepSort

scDeepSort performs reference-free cell-type annotation of single-cell RNA sequencing (scRNA-seq) data using a deep learning approach based on a modified graph neural network (GNN) to identify cellular identities for biological and disease studies.

Key Features:

• Reference-Free Annotation: Operates without predefined marker genes or external RNA-seq reference profiles, annotating cells from intrinsic scRNA-seq transcriptomic data.
• Graph Neural Network (GNN) Model: Implements a modified GNN deep learning model that constructs weighted graphs from single-cell transcriptomes to capture relationships between cells.
• High Accuracy and Scalability: Validated on datasets of 764,741 cells across 56 human and 32 mouse tissues and evaluated on external datasets of 126,384 human and 134,604 mouse cells, demonstrating superior labeling accuracy compared to reference-dependent methods.
• Broad Applicability: Enables annotation at single-cell resolution to explore cellular heterogeneity and study biological processes such as disease pathogenesis and progression.

Scientific Applications:

• Cell Type Identification: Provides robust identification of cell types within complex tissues from scRNA-seq data.
• Biological Insights: Facilitates uncovering mechanisms underlying cellular heterogeneity and biological processes without relying on reference datasets.
• Disease Research: Supports studies of disease pathogenesis and progression at single-cell resolution.

Methodology:

Uses a modified graph neural network (GNN) deep learning model that constructs weighted graphs from single-cell transcriptomes to perform reference-free cell-type annotation without using predefined marker genes or external RNA-seq profiles.