STAMarker: Determining spatial domain-specific variable genes with saliency maps in deep learning

• DOI: 10.1101/2022.11.07.515535
• Publication Date: 2022-11-08T23:55:15Z
• Authors: Chihao Zhang; Kangning Dong; Kazuyuki Aihara; Luonan Chen; Shihua Zhang
• Keywords: Spatial contextual awareness

Abstract Spatial transcriptomics characterizes gene expression profiles while retaining the information of spatial context, providing an unprecedented opportunity to understand cellular systems. One essential tasks in such data analysis is determine spatially variable genes (SVGs), which demonstrate patterns. Existing methods only consider individually and fail model inter-dependence genes. To this end, we present analytic tool STAMarker for robustly determining domain-specific SVGs with saliency maps deep learning. a three-stage ensemble framework consisting graphattention autoencoders, multilayer perceptron (MLP) classifiers, map computation by backpropagated gradient. We illustrate effectiveness compare it three competing on four transcriptomic generated various platforms. considers all at once more robust when dataset very sparse. could identify characterizing domains enable in-depth region interest tissue section.