A graph neural network-based spatial multi-omics data integration method for deciphering spatial domains.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Spatial sequencing technologies allow simultaneous transcriptomic and epigenomic profiling within tissue slices.
• Understanding cellular microenvironments requires integrative analysis of spatial multi-omics data.
• Current methods lack effective approaches for integrating diverse spatial multi-omics datasets.

Purpose of the Study:

• To develop an effective computational model for integrating spatial multi-omics data.
• To enhance the identification of spatial domains and improve data quality in spatial multi-omics studies.
• To address the limitations of existing methods in analyzing complex spatial multi-omics datasets.

Main Methods:

• Proposed SpaMI, a graph neural network-based model.
• Utilized a contrastive learning strategy for feature extraction from individual omics data.
• Employed an attention mechanism for integrating features across different omics layers.
• Applied the model to simulated data and real spatial epigenome-transcriptome and transcriptome-proteome datasets.

Main Results:

• SpaMI demonstrated superior performance in identifying spatial domains compared to state-of-the-art methods.
• The model effectively denoises spatial multi-omics data.
• Successful application to both simulated and real-world spatial multi-omics datasets, including epigenome-transcriptome and transcriptome-proteome data.

Conclusions:

• SpaMI provides a powerful and effective approach for the integrative analysis of spatial multi-omics data.
• The model advances the understanding of cellular microenvironments by enabling robust analysis of integrated spatial omics profiles.
• SpaMI represents a significant improvement for spatial domain identification and data denoising in multi-omics studies.