Abstract PD11-05: Inferring Spatial Genomic Expression profiles from H&E-Stained Histology Using Weakly Supervised Deep Learning in Breast Cancer

Key Points

• This research aims to develop an AI model to infer spatial gene expression from H&E-stained images in breast cancer.
• Trained a weakly supervised deep learning model using a public dataset (TCGA)
• Utilized bulk RNA-seq data paired with H&E-stained whole slide images
• Evaluated prediction accuracy using Pearson’s correlation on a validation dataset
• Compared predicted spatial expressions with immunohistochemistry data and spatial transcriptomics
• Predicted expression of over 2,500 genes with a Pearson correlation greater than 0.6
• Notable genes like ESR1 and AURKA showed high correlation with actual RNA-seq data
• Predicted patterns aligned with IHC-stained tissue sections and spatial transcriptomics data