1160-DeepSpaCE: for Super-Resolution Spatial Transcriptomics

This paper introduces DeepSpaCE, a deep-learning tool designed to enhance spatial transcriptomics by predicting gene-expression profiles from standard histological images. Traditional spatial gene mapping is often hindered by high experimental costs and technical errors, such as permeabilization issues that leave gaps in tissue data. By utilizing convolutional neural networks, this method enables super-resolution imaging of gene activity and the imputation of missing data in consecutive tissue sections without additional experiments. Researchers validated the model using human breast cancer samples, demonstrating its ability to accurately identify critical markers like ESR1 and the invasion marker SPARC. Furthermore, the integration of semi-supervised learning allows the model to maintain high accuracy even with limited sample sizes. Ultimately, this technology offers a cost-effective way for biologists to uncover hidden histological features and functional boundaries in complex tissues.References:Monjo, T., Koido, M., Nagasawa, S. et al. Efficient prediction of a spatial transcriptomics profile better characterizes breast cancer tissue sections without costly experimentation. Sci Rep 12, 4133 (2022). https://doi.org/10.1038/s41598-022-07685-4前往小宇宙评论区与主播互动