Generation of super-resolution images from barcode-based spatial transcriptomics by deep image prior

• Published in: Cell reports methods Vol. 5; no. 1; p. 100937
• Main Authors: Park, Jeongbin, Cook, Seungho, Lee, Dongjoo, Choi, Jinyeong, Yoo, Seongjin, Bae, Sungwoo, Im, Hyung-Jun, Lee, Daeseung, Choi, Hongyoon
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 27.01.2025; Elsevier
• Subjects: Algorithms; deep image prior; dense matrix; gene expression; Gene Expression Profiling - methods; high-resolution image; Humans; Image Processing, Computer-Assisted - methods; spatial transcriptomics; Transcriptome - genetics; CP: Systems biology; dense matrix; spatial transcriptomics; CP: Imaging; deep image prior; high-resolution image; gene expression
• ISSN: 2667-2375; 2667-2375
• DOI: 10.1016/j.crmeth.2024.100937

Spatially resolved transcriptomics (ST) has revolutionized the field of biology by providing a powerful tool for analyzing gene expression in situ. However, current ST methods, particularly barcode-based methods, have limitations in reconstructing high-resolution images from barcodes sparsely distributed in slides. Here, we present SuperST, an algorithm that enables the reconstruction of dense matrices (higher-resolution and non-zero-inflated matrices) from low-resolution ST libraries. SuperST is based on deep image prior, which reconstructs spatial gene expression patterns as image matrices. Compared with previous methods, SuperST generated output images that more closely resembled immunofluorescence images for given gene expression maps. Furthermore, we demonstrated how one can combine images created by SuperST with computer vision algorithms. In this context, we proposed a method for extracting features from the images, which can aid in spatial clustering of genes. By providing a dense matrix for each gene in situ, SuperST can successfully address the resolution and zero-inflation issue. [Display omitted] •SuperST enhances spatial transcriptomics resolution using deep image prior•It outperforms other methods in generating dense gene expression matrices•Reduced parameter sensitivity enables reliable performance across datasets Enhancing the resolution of barcode-based spatial transcriptomics can deepen our understanding of biological phenomena using both existing and newly generated data. We propose a method, SuperST, that aims to overcome the limitations of spatially resolved transcriptomics (ST) methods that rely on segmentation and/or exhibit parameter-dependent performance. SuperST utilizes an end-to-end convolutional neural network and is robust to parameters such as image size, learning epochs, or learning rate. To enhance usability, we integrate Visium-specific diffusion features and offer user-friendly Python-based implementations. Park et al. present SuperST, an algorithm that enhances spatial transcriptomics by reconstructing high-resolution, dense gene expression matrices from low-resolution raw data. Using deep image prior, SuperST produces spatial patterns that closely resemble immunofluorescence images. The method overcomes limitations in resolution and zero inflation, enabling applications based on computer vision.