The rapid development of spatial transcriptomics allows the measurement of RNA abundance at a high spatial resolution making it possible to simultaneously profile gene expression spatial locations of

The rapid advancement of spatial transcriptomics has allowed for the measurement of RNA abundance with high spatial resolution. This technology makes it possible to simultaneously analyze gene expression, the spatial locations of cells or spots, and corresponding histology images. This opens up the possibility of predicting gene expression from histology images, which are relatively easy and inexpensive to obtain. While several methods have been developed for this purpose, they have not fully captured the internal relationships of the 2D vision features or the spatial dependency between spots.

In this study, we introduce Hist2ST, a deep learning-based model that can predict RNA-seq expression from histology images. The model works by cropping the corresponding histology image into an image patch around each sequenced spot. These image patches are then fed into a convolutional module to extract 2D vision features. Additionally, the model captures spatial relations with the whole image and neighboring patches using Transformer and graph neural network modules, respectively. The learned features are used to predict gene expression by following the zero-inflated negative binomial distribution.

To address the issue of limited spatial transcriptomics data, we employ a self-distillation mechanism for efficient learning of the model. Through comprehensive tests on cancer and normal datasets, Hist2ST has demonstrated superior performance in terms of both gene expression prediction and spatial region identification compared to existing methods. Furthermore, pathway analyses have indicated that our model can preserve important biological information.

In summary, Hist2ST offers a powerful tool for generating spatial transcriptomics data from histology images. This allows for a deeper understanding of the molecular signatures of tissues and holds great potential for advancing our knowledge of gene expression patterns in various biological contexts