scGraphDap: Integrating Functional State Pseudo-Labels and Graph Structure Learning for Robust Cell Type Annotation...

Area of Science:

• Computational biology
• Cancer research
• Bioinformatics

Background:

• The tumor microenvironment (TME) is crucial for cancer progression, involving complex cell-cell interactions.
• Analyzing cell-cell communication from single-cell RNA sequencing (scRNA-seq) data is challenging due to data limitations.

Purpose of the Study:

• To develop a novel graph neural network framework, scGraphDap, for improved cell type annotation and cell-cell communication (CCC) inference.
• To enhance the understanding of TME dynamics and identify potential therapeutic targets.

Main Methods:

• scGraphDap integrates functional state pseudo-labels (pathway activity scores) with graph structure learning.
• A graph domain adaptation module improves cross-patient generalization.
• The framework leverages pathway activity scores to optimize cell-cell graphs for functional proximity.

Main Results:

• scGraphDap achieved an average accuracy of 82.82% in cell type annotation and CCC inference across 38,667 cells from 15 patients.
• The method successfully identified disease-specific gene interactions, such as STAT3-CD274 in breast cancer.
• The framework demonstrated robust performance in uncovering functional proximity beyond simple geometric similarity.

Conclusions:

• scGraphDap provides a unified and effective framework for TME analysis using non-spatial scRNA-seq data.
• The approach offers valuable insights into cancer progression mechanisms and aids in discovering novel therapeutic targets.