New GNN predicts gene expression from pathology images across scales

Researchers have developed a novel Multi-Scale Graph Neural Network (GNN) capable of predicting gene expression at various resolutions—bulk, spatial, and single-cell—directly from Whole-Slide Images (WSIs). This model leverages a Pathology Foundation Model (UNI2-h) to extract morphological features and constructs spatial graphs at both patch and cell levels. Trained on data from The Cancer Genome Atlas Program, Visium, and Xenium across lung, kidney, and breast tissues, the GNN demonstrates accuracy in predicting slide-level gene expression and moderate success at patch and single-cell levels. The predictions also show utility in downstream clinical tasks such as survival analysis and tumor classification. AI

IMPACT This model could enable more cost-effective and scalable analysis of gene expression from readily available pathology images, aiding disease research.

RANK_REASON The item is an academic paper detailing a new machine learning model for biological data analysis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• kidney
• lung
• Multi-Scale Graph Neural Network
• Pathology Foundation Model
• single-cell RNA-seq
• spatial transcriptomics
• The Cancer Genome Atlas Program
• UNI2-h
• Xenium

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →