Single cell thesis record (Part12) -- unsupervised spatial embedded deep representation of spatial transcriptomics

Learning notes , For reference only , If there is a mistake, it must be corrected
Authors:Fu Huazhu, XU Hang,Chen Jinmiao
Year：2021
Key words: spatial transcriptomics; graph convolutional network; gene expression; deep learning

---

Unsupervised Spatial Embedded Deep Representation of Spatial Transcriptomics

Abstract

Spatial transcriptomics enable us to dissect tissue heterogeneity and map out inter-cellular communications. Optimal integration of transcriptomics data and associated spatial information is essential towards fully exploiting the data. We present SEDR, an unsupervised spatially embedded deep representation of both transcript and spatial information. The SEDR pipeline uses a deep autoencoder to construct a low-dimensional latent representation of gene expression, which is then simultaneously embedded with the corresponding spatial information through a variational graph autoencoder. We applied SEDR on human dorsolateral prefrontal cortex data and achieved better clustering accuracy, and correctly retraced the prenatal cortex development order with trajectory analysis. We also found the SEDR representation to be eminently suited for batch integration. Applying SEDR to human breast cancer data, we discerned heterogeneous sub-regions within a visually homogenous tumor region, identifying a tumor core with pro-inflammatory microenvironment and an outer ring region enriched with tumor associated macrophages which drives an immune suppressive microenvironment.

Overview of SEDR.

SEDR learns a gene representation in a low-dimensional latent space with jointly embedded spatial information (Figure 1). Given spatial transcriptomics data, SEDR first learns a nonlinear mapping from the gene expression space to a low-dimensional feature space using a deep autoencoder network. Simultaneously, a variational graph autoencoder is utilized to aggregate the gene representation with the corresponding spatial neighborhood relationships to produce a spatial embedding. Then, the gene representation and spatial embedding are concatenated to form the final latent representation used to reconstruct the gene expression. Thereafter, an unsupervised deep clustering method is employed to enhance the compactness of the learned latent representation. This iterative deep clustering generates a form of soft clustering that assigns cluster-specific probabilities to each cell, leveraging on the inferences between cluster-specific and cell-specific representation learning. Finally, the learned latent representation can be applied towards various analysis tasks.

Overview of SEDR. SEDR learns a low-dimensional latent representation of gene expression embedded with spatial information by jointly training a deep autoencoder and a variational graph autoencoder. The low-dimensional embedding produced by SEDR can be used for downstream visualization, cell clustering, trajectory inference, and batch effect correction.