Enhancing Spatial Transcriptomics Analysis by Integrating Image-Aware Deep Learning Methods.

Area of Science:

• Biomedical Research
• Computational Biology
• Cancer Research

Background:

• Spatial transcriptomics (ST) profiles cell transcription in tissue context.
• Current ST analysis underutilizes morphological data, focusing primarily on gene expression.
• Understanding spatial heterogeneity in cancer requires integrating diverse data types.

Purpose of the Study:

• To develop a novel method integrating ST and histopathology images for comprehensive cancer tissue analysis.
• To leverage morphological features from histology images alongside gene expression data.
• To improve the detection of biologically meaningful patterns in aggressive cancers like glioblastoma and triple-negative breast cancer.

Main Methods:

• Utilized a ResNet-based deep learning model to extract morphological features from whole-slide histology images.
• Combined PCA-reduced vectors from image analysis and ST gene expression data.
• Applied Louvain clustering for image-aware feature discovery.

Main Results:

• Image-aware clustering successfully identified key histopathological features such as fibrosis and necrosis.
• Improved definition of EGFR-rich areas was achieved through the integrated approach.
• The combinatorial method revealed critical histopathological characteristics missed by gene-expression-only analyses.

Conclusions:

• Integrating spatial transcriptomics with histopathology image data offers a more comprehensive understanding of cancer tissue biology.
• This novel approach enhances the discovery of subtle yet significant biological patterns.
• The method holds promise for advancing the analysis of aggressive cancers and personalized medicine.