Marker Gene-Guided Graph Neural Networks for Enhanced Spatial Transcriptomics Clustering.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial Transcriptomics (ST) enables cell relationship analysis within tissue context.
• Current ST clustering methods lack domain knowledge integration, such as marker genes.
• Marker gene information can improve cell embedding and clustering accuracy.

Purpose of the Study:

• To introduce a novel approach, Marker Gene-Guided Graph Neural Networks (MGGNN), for enhanced spatial transcriptomics clustering.
• To leverage marker gene information to improve cell embedding and clustering outcomes in ST data.

Main Methods:

• Developed a Graph Neural Network (GNN) based contrastive learning framework.
• Fine-tuned the GNN model using limited marker gene expression data for spot labeling.
• Evaluated MGGNN performance through simulations and experiments on two real-world ST datasets.

Main Results:

• MGGNN demonstrated superior performance compared to state-of-the-art clustering methods.
• The integration of marker genes significantly enhanced the learning of cell embeddings.
• The model achieved improved clustering outcomes on both simulated and real-world ST data.

Conclusions:

• Marker Gene-Guided Graph Neural Networks (MGGNN) offer a powerful new method for spatial transcriptomics analysis.
• Incorporating domain knowledge, like marker genes, is crucial for advancing ST data interpretation.
• MGGNN provides a robust and effective solution for accurate cell clustering in spatial transcriptomics.