HierarchicalDAEW: Domain-Aware Edge-Weighted Graph Convolution with Evidential Uncertainty for Multi-Section Spatial Gene Expression Prediction from H&E Histology

2026-07-23 · Source: Machine Learning · Field: Science & Research — Life Sciences & Biology, Health & Medical Research, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

HierarchicalDAEW is a novel dual-graph architecture designed to predict spatially resolved gene expression directly from H&E histology, addressing the high cost and technical demands of current spatial transcriptomics assays. This model incorporates a Domain-Aware Edge-Weighted convolutional operator on a spot graph, which learns distinct projections for inter-domain, intra-domain, and boundary edges derived from Leiden clustering, explicitly treating tissue heterogeneity as a structural signal. A second gene-level graph integrates protein-protein interaction priors from STRING-DB with tissue-specific co-expression via learned attention gating. Crucially, HierarchicalDAEW employs evidential uncertainty estimation, yielding better calibrated confidence intervals than Monte Carlo dropout. It achieved the strongest correlation with ground-truth expression across six human Visium sections (breast, colorectal, prostate, cerebellar tissue) against thirteen published baselines, with its components confirmed by ablation studies.

Key takeaway

For clinical researchers evaluating spatial gene expression prediction methods, HierarchicalDAEW offers a robust solution by providing superior accuracy and reliably calibrated uncertainty estimates. You should consider its dual-graph architecture, which explicitly models tissue heterogeneity and integrates protein interaction data, to enhance prediction quality. Its evidential uncertainty estimation is critical for identifying low-confidence predictions, ensuring that you can direct pathologist review effectively before clinical action.

Key insights

HierarchicalDAEW predicts spatial gene expression from H&E histology using a dual-graph architecture and evidential uncertainty for reliability.

Principles

• Explicitly model tissue heterogeneity as a structural signal.
• Fuse protein-protein interaction priors with co-expression data.
• Evidential uncertainty provides superior confidence calibration.

Method

HierarchicalDAEW utilizes a spot graph with Domain-Aware Edge-Weighted convolution and a gene-level graph that fuses STRING-DB priors with tissue-specific co-expression via attention gating for prediction.

In practice

• Use Leiden clustering to define tissue domains.
• Integrate STRING-DB for protein-protein interaction priors.
• Flag low-confidence predictions for pathologist review.

Topics

• Spatial Transcriptomics
• Gene Expression Prediction
• H&E Histology
• Graph Convolutional Networks
• Evidential Uncertainty
• Protein-Protein Interaction

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• COAST: Context-Aware Differential Learning for Gene Expression Prediction in Spatial Transcriptomics — COAST predicts spatial gene expression from H&E images by combining absolute and differential learning with context-aware Transformer encoding.
• Virtual 3D H&E Staining from Phase-contrast Back-illumination Interference Tomography — A novel dataset and GAN framework enable accurate 3D virtual H&E staining from label-free Back-illumination Interference Tomography.
• Prior-Anchored Debiasing for Long-Tailed Multi-Organ Pathology Report Generation — PriOrGen mitigates long-tail biases in multi-organ pathology report generation using visual and textual anchoring.
• Evidential learning driven Breast Tumor Segmentation with Stage-divided Vision-Language Interaction — TextBCS enhances breast tumor segmentation in MRI by combining text prompts with evidential learning to address low contrast and blurred boundaries.
• Systematically decoding pathological morphologies and molecular profiles with unified multimodal embedding — Multi-Embed unifies morphological and molecular data for superior disease understanding and prediction.
• Knowledge Graphs and Reasoning LLMs for Finding Simple Yet Effective Transcriptomic Perturbation Predictors — Simple K-nearest neighbour models leveraging knowledge graphs, enhanced by RL-trained LLMs, offer competitive transcriptomic perturbation prediction.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.