Predicting spatially resolved gene expression via tissue morphology using adaptive spatial GNNs.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics offers insights into tissue architecture and molecular mechanisms but is limited by high costs.
• Histological images are routinely generated and more affordable, presenting an opportunity for alternative gene expression analysis.

Purpose of the Study:

• To develop a scalable method for predicting spatial gene expression from histological images.
• To leverage morphological information in histology for molecular decoding of tissue complexity.

Main Methods:

• A graph neural network (GNN) framework was developed to predict spatial gene expression.
• The model analyzes histological images to infer gene activity patterns.

Main Results:

• The GNN framework demonstrated improved prediction performance over existing state-of-the-art methods.
• Experiments were conducted on two independent breast cancer cohorts, validating the model's efficacy.
• The model successfully delineated spatial domains of biological significance.

Conclusions:

• Predicting spatial gene expression from histology images is a viable and scalable alternative to costly spatial transcriptomics.
• The developed GNN framework offers a powerful tool for dissecting tissue complexity and advancing biomedical research.