Identifying spatial domain by adapting transcriptomics with histology through contrastive learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics offers cell/spot resolution gene expression with spatial and histological context.
• Accurate spatial domain identification is crucial for downstream analysis.
• Existing methods struggle to fully integrate gene expression and image data.

Purpose of the Study:

• To develop a novel method, ConGI, for accurate spatial domain identification in spatial transcriptomics.
• To leverage contrastive learning for effective integration of gene expression and histopathological image data.
• To improve downstream analysis tasks by learning coupled representations.

Main Methods:

• Propose ConGI, a method using contrastive learning to adapt gene expression with histopathological images.
• Design three contrastive loss functions for learning common representations between gene expression and image modalities.
• Apply learned representations for clustering spatial domains on tumor and normal datasets.

Main Results:

• ConGI demonstrates superior performance in spatial domain identification compared to existing methods.
• The learned representations from ConGI are effective for various downstream tasks.
• Successful clustering of spatial domains on both tumor and normal spatial transcriptomics datasets.

Conclusions:

• ConGI effectively integrates spatial transcriptomics gene expression and histology images using contrastive learning.
• The method achieves state-of-the-art performance in spatial domain identification.
• Learned representations enhance downstream analyses like trajectory inference, clustering, and visualization.