STGNNks: Identifying cell types in spatial transcriptomics data based on graph neural network, denoising...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics integrates spatial location, tissue morphology, and gene expression data.
• Analyzing this integrated data is crucial for understanding cell biology within its tissue context.

Purpose of the Study:

• To develop an innovative spatial clustering method, STGNNks, for enhanced spatial transcriptomics data analysis.
• To accurately identify spatial domains and infer biological processes from high-resolution transcriptomic data.

Main Methods:

• STGNNks combines graph neural networks, denoising auto-encoders, and k-sums clustering.
• It preprocesses spatial transcriptomics data, constructs a hybrid adjacency matrix, and learns embedding features using a graph convolutional network.
• Learned features are dimension-reduced via a ZINB-based auto-encoder, followed by k-sums clustering for domain identification.

Main Results:

• STGNNks significantly outperformed five other spatial clustering methods (CCST, Seurat, stLearn, Scanpy, SEDR) across multiple internal and external validation metrics.
• The method successfully identified spatially variable and differentially expressed genes in unlabeled datasets.
• Trajectory inference revealed disease progression patterns in human breast cancer data.

Conclusions:

• STGNNks offers an efficient and powerful approach for spatial transcriptomics data analysis.
• The method has the potential to improve disease diagnosis and therapy by providing deeper biological insights.
• The STGNNks code is publicly available for broader research application.