Related Experiment Video
Updated: Sep 10, 2025
09:19
Spatial Profiling of Protein and RNA Expression in Tissue: An Approach to Fine-Tune Virtual Microdissection
Published on: July 6, 2022
5.0K
DANet: spatial gene expression prediction from H&E histology images through dynamic alignment
Yulong Wu1, Jin Xie1, Jing Nie2,3
1School of Big Data and Software Engineering, Chongqing University, Chongqing 400044, China.
Briefings in Bioinformatics
|August 20, 2025
Summary
Predicting spatial gene expression from histology images is faster and cheaper than sequencing. Our novel method improves accuracy by capturing local details and gene correlations using dense structures and state space models.
Area of Science:
- Computational pathology
- Bioinformatics
- Machine learning in medicine
Background:
- Spatial gene expression prediction from histology images can accelerate research and understanding of tissue architecture and disease.
- Current methods struggle with fine-grained local feature extraction and modeling gene-gene correlations.
- Aligning heterogeneous modalities in bimodal contrastive learning is a significant challenge.
Purpose of the Study:
- To develop a novel method for accurate spatial gene expression prediction from Hematoxylin and Eosin histology images.
- To enhance the capture of local image features and improve gene-gene correlation modeling.
- To address challenges in dynamic and efficient alignment of heterogeneous modalities in contrastive learning.
Main Methods:
- Introduced a dense connective structure for efficient feature reuse and local feature mining.
- Utilized state space models to capture dependencies within 1D gene expression data for improved gene-gene correlation modeling.
- Designed the Residual Kolmogorov-Arnold Network (RKAN) with a learnable activation function for dynamic bimodal mapping and alignment during contrastive training.
Main Results:
- Demonstrated significant improvements in spatial gene expression prediction accuracy.
- Showcased the effectiveness of the dense connective structure in capturing local image details.
- Validated the ability of state space models and RKAN to model gene-gene correlations and align modalities effectively.
- Achieved superior performance on two public datasets (GSE240429 and HER2+) compared to existing methods.
Conclusions:
- The proposed method significantly enhances spatial gene expression prediction from histology images.
- The novel architecture effectively addresses limitations in feature extraction, gene-gene correlation modeling, and multimodal alignment.
- This approach offers a more efficient and accurate alternative to traditional gene expression sequencing for understanding tissue biology and disease.
