Benchmarking the translational potential of spatial gene expression prediction from histology

Chan, Adam S.; Wang, Chuhan; Fu, Xiaohang; Ghazanfar, Shila; Kim, Jinman; Patrick, Ellis; Yang, Jean YH

doi:10.1101/2023.12.12.571251

“…Next, we note that the performance of our method on the top five biologically meaningful genes ( Figure 3c ) had the highest correlation compared to other methods, with GNAS (r = 0.42), FASN (r = 0.42), SCD (r = 0.34), MYL12B (r = 0.32), and CLDN4 (r = 0.32), indicating one should examine correlation among genes with meaningful biological signal. Thus, with the limitation of the PCC values in mind, we examined more meaningful gene characteristics metrics, as discussed in our recent benchmarking work 15 , where we evaluated based on top 10% of HVGs and top 20 SVGs. Figures 3d and 3e, and Supplementary Figure 6 show that our method provided the highest PCC in HVGs (0.20) and SVGs (0.27), and also SSIM in HVGs (0.17) and SVGs (0.26).…”

Section: Resultsmentioning

confidence: 99%

“…We describe the settings for other spot-based methods below, and more details may be found in our benchmarking work 15 :…”

Section: Methodsmentioning

confidence: 99%

“…GHIST can be easily adapted to predict spot-based SGE. Here, we use an existing benchmarking framework proposed for spot-base methods 15 to establish the extensibility of our method. Using the HER2ST dataset 23 , we first evaluated the spot-based SGE predictions with metrics commonly used in the literature such as Pearson Correlation Coefficient (PCC) and Structural Similarity Index (SSIM) across all genes for each method.…”

Section: Ghist Is Extendable To Spot-based Technologies and Outperfor...mentioning

confidence: 99%

“…[B] Evaluation measures Concordance between predicted and measured gene expression -We followed the strategies described in our benchmarking work 15 to assess the performance of GHIST on HER2ST data. Briefly, the evaluation methods used include PCC and SSIM.…”

Section: Evaluation Study 2: Comparison Study With Other Spot-based M...mentioning

confidence: 99%

“…Super-resolved pixels, however, potentially contain a mixture of cells with background areas, and are thus not at true single-cell resolution. Our previous benchmarking study 15 on the existing methods revealed a weak translational potential, which is likely due to the limited information in the predictions. In these results, expressions in each spot describe a mixture of cells and a mixture of cell types, which contain limited power as the spot-based SGE expressions are not at single-cell resolution.…”

Section: Introductionmentioning

confidence: 96%

See 3 more Smart Citations

Spatial gene expression at single-cell resolution from histology using deep learning with GHIST

Fu,

Cao,

Bian

et al. 2024

Preprint

Self Cite

1

0

View full text Add to dashboard Cite

The increased use of spatially resolved transcriptomics provides new biological insights into disease mechanisms. However, the high cost and complexity of these methods are barriers to broad clinical adoption. Consequently, methods have been created to predict spot-based gene expression from routinely-collected histology images. Recent benchmarking showed that current methodologies have limited accuracy and spatial resolution, constraining translational capacity. Here, we introduce GHIST, a deep learning-based framework that predicts spatial gene expression at single-cell resolution by leveraging subcellular spatial transcriptomics and synergistic relationships between multiple layers of biological information. We validated GHIST using public datasets and The Cancer Genome Atlas data, demonstrating its flexibility across different spatial resolutions and superior performance. Our results underscore the utility ofin silicogeneration of single-cell spatial gene expression measurements and the capacity to enrich existing datasets with a spatially resolved omics modality, paving the way for scalable multi-omics analysis and new biomarker discoveries.

show abstract

Gene expression prediction from histology images via hypergraph neural networks

Li,

Zhang,

Wang

et al. 2024

Briefings in Bioinformatics

1

0

View full text Add to dashboard Cite

Spatial transcriptomics reveals the spatial distribution of genes in complex tissues, providing crucial insights into biological processes, disease mechanisms, and drug development. The prediction of gene expression based on cost-effective histology images is a promising yet challenging field of research. Existing methods for gene prediction from histology images exhibit two major limitations. First, they ignore the intricate relationship between cell morphological information and gene expression. Second, these methods do not fully utilize the different latent stages of features extracted from the images. To address these limitations, we propose a novel hypergraph neural network model, HGGEP, to predict gene expressions from histology images. HGGEP includes a gradient enhancement module to enhance the model’s perception of cell morphological information. A lightweight backbone network extracts multiple latent stage features from the image, followed by attention mechanisms to refine the representation of features at each latent stage and capture their relations with nearby features. To explore higher-order associations among multiple latent stage features, we stack them and feed into the hypergraph to establish associations among features at different scales. Experimental results on multiple datasets from disease samples including cancers and tumor disease, demonstrate the superior performance of our HGGEP model than existing methods.

show abstract

Gene Expression Prediction from Histology Images via Hypergraph Neural Networks

Li,

Zhang,

Wang

et al. 2024

Preprint

0

View full text Add to dashboard Cite

Spatial transcriptomics reveals the spatial distribution of genes in complex tissues, providing crucial insights into biological processes, disease mechanisms, and drug development. The prediction of gene expression based on cost-effective histology images is a promising yet challenging field of research. Existing methods for gene prediction from histology images exhibit two major limitations. First, they ignore the intricate relationship between cell morphological information and gene expression. Second, these methods do not fully utilize the different latent stages of features extracted from the images. To address these limitations, we propose a novel hypergraph neural network model, HGGEP, to predict gene expressions from histology images. HGGEP includes a gradient enhancement module to enhance the model’s perception of cell morphological information. A lightweight backbone network extracts multiple latent stage features from the image, followed by attention mechanisms to refine the representation of features at each latent stage and capture their relations with nearby features. To explore higher-order associations among multiple latent stage features, we stack them and feed into the hypergraph to establish associations among features at different scales. Experimental results on multiple datasets from disease samples including cancers and tumor disease, demonstrate the superior performance of our HGGEP model than existing methods.Key PointsWe develop a novel histology image-based gene prediction model named HGGEP, which demonstrates high accuracy and robust performance.To reveal the intricate relationship between cell morphology and gene expression in images, we propose a gradient enhancement module, which effectively improves the model’s capability in perceiving cell morphology in images.HGGEP includes a hypergraph module that efficiently models higher-order associations among features across multiple latent stages, resulting in significant performance improvement.

show abstract

Benchmarking the translational potential of spatial gene expression prediction from histology

Cited by 4 publications

References 59 publications

Spatial gene expression at single-cell resolution from histology using deep learning with GHIST

Spatial gene expression at single-cell resolution from histology using deep learning with GHIST

Gene expression prediction from histology images via hypergraph neural networks

Gene Expression Prediction from Histology Images via Hypergraph Neural Networks

Contact Info

Product

Resources

About