Researchers from the Chinese University of Hong Kong have introduced an innovative approach called MultiGATE, designed to integrate complex spatial multi-omics data. Spatial multi-omics combines information about where specific molecules are located in tissues with what genes or proteins they represent. By linking transcriptomics (RNA sequencing data) with other molecular features like proteins and epigenetic markers, scientists can better understand how cells interact and regulate gene activity in their natural tissue environments.

Overview of MultiGATE

MultiGATE is a two-level graph attention auto-encoder designed for spatial multi-omics analysis. It extracts the latent embeddings of the pixels/spots in spatial multi-omics data, while simultaneously incorporating the regulatory relationship of the cross-modality features through the cross-modality attention mechanism and the spatial relationship of the pixels/spots through the within-modality attention mechanism. In addition to reconstruction loss, a CLIP contrastive loss aligns embeddings across modalities. MultiGATE yields (i) latent representations of pixels for clustering and visualization and (ii) cross-modality attention scores for cross-modality regulatory inference.

MultiGATE uses a two-level graph attention auto-encoder, a type of artificial intelligence model that identifies patterns in complex datasets. This model simultaneously captures the spatial arrangement of cells and predicts how different molecular layers influence one another. Essentially, MultiGATE provides a deeper look at transcriptional regulation, which governs how genes are turned on or off across different tissue regions.

The research team tested MultiGATE on multiple tissue types and technological platforms, showing that it improves both data integration and biological insight. By uncovering hidden relationships between gene expression and other molecular signals, this tool enhances our understanding of how different regions of tissue coordinate their activity, which could ultimately advance disease research and biomarker discovery.

Availability – https://github.com/cuhklinlab/MultiGATE