Sunday 02 February 2025
A team of researchers has developed a novel approach to predicting gene expression in tumor tissues using whole slide imaging (WSI) data. The method, known as MERGE, combines spatial and feature space clustering techniques with graph neural networks to accurately predict gene expression patterns.
The researchers used the ST-Net dataset, which contains WSI images of breast cancer tumors along with corresponding gene expression data. They first applied a smoothing technique called SPCS to remove noise and improve the quality of the imaging data.
Next, they used a ResNet18-based patch encoder to extract features from the smoothed WSI images. These features were then fed into a graph neural network that was trained to predict gene expression patterns.
The researchers found that MERGE outperformed two baseline methods, BLEEP and TRIPLEX, in predicting gene expression patterns for two cancer-relevant genes: FASN and GNAS. They also found that the combination of spatial and feature space clustering techniques improved the accuracy of gene expression prediction.
One of the key innovations of MERGE is its ability to capture complex relationships between genes and tissue morphology. By using graph neural networks, the method can learn to identify patterns in the WSI images that are associated with specific gene expressions.
The researchers believe that MERGE has the potential to be used in clinical settings to help diagnose and treat cancer more effectively. They plan to continue developing the method and testing it on larger datasets.
In summary, a new approach called MERGE has been developed that uses spatial and feature space clustering techniques combined with graph neural networks to predict gene expression patterns from whole slide imaging data. The method outperforms existing approaches in predicting gene expression patterns for two cancer-relevant genes and has the potential to be used in clinical settings to help diagnose and treat cancer more effectively.
Cite this article: “MERGE: A Novel Approach to Predicting Gene Expression from Whole Slide Imaging Data”, The Science Archive, 2025.
Gene Expression, Whole Slide Imaging, Merge, Graph Neural Networks, Spatial Clustering, Feature Space Clustering, Breast Cancer, Spcs, Resnet18, Wsi Images
