همایش ملی بیوانفورماتیک ایران

صفحه اصلی / 4th international edition and 13th Iranian Conference on Bioinformatics

Uncovering Disrupted Cell-Cell Interactions in Alzheimer's Disease Using Variational Graph Autoencoders on Single-Cell Spatial Transcriptomics Data from the Human Middle Temporal Gyrus

نویسندگان :

Zahra Bayat¹ Alireza Fotuhi Siahpirani²

1- Laboratory of Bioinformatics and Computational Genomics (LBCG), Department of Bioinformatics, Institute of Biochemistry and Biophysics (IBB), University of Tehran, Tehran, Iran 2- Laboratory of Bioinformatics and Computational Genomics (LBCG), Department of Bioinformatics, Institute of Biochemistry and Biophysics (IBB), University of Tehran, Tehran, Iran

کلمات کلیدی :

Middle Temporal Gyrus (MTG)،Alzheimer's Disease (AD)،Cell Cell interaction،Single-Cell Spatial Transcriptome،Variational Graph Autoencoder (VGAE)

چکیده :

The human middle temporal gyrus (MTG) is a susceptible brain region in the early stages of Alzheimer's disease (AD), yet the molecular mechanisms driving this regional vulnerability remain poorly understood. Understanding the intricate cell-cell interactions (CCIs) in this region is crucial for elucidating the pathophysiology of AD. Recent advances in single-cell spatial transcriptomics and deep learning have created new opportunities to explore CCIs in unprecedented detail. Variational graph autoencoders (VGAEs), a state-of-the-art deep generative model for graph-structured data, enable the encoding of complex interactions between cells while preserving their spatial and molecular context. Chen and colleagues (Chen et al. 2020) have used 10x Visium platform to measure spatial transcriptomics in AD and control MTG samples. The dataset was obtained from GEO (GSE220442) and analyzed with Seurat. Data were log-normalized, scaled, and integrated using reciprocal principal component analysis (RPCA) and the Louvain algorithm was used for dimensionality reduction and clustering. To employ the VGAE model for inferring cell-cell interactions, we used the DeepLinc pipline (Li & Yang 2022) with two graph convolutional layers, which requires two input files: the adjacency matrix of a cell-cell interaction graph (A) and a gene expression matrix (X) as features of the nodes in A. The adjacency matrix was created by using the K-nearest neighbors (KNN) algorithm to find the three closest neighbors for each cell based on geometric closeness, with a distance threshold applied to consider only nearby cells as direct contacts. The latent representations learned by DeepLinc were used to infer known and novel CCIs. Cell-type specific differentially expressed genes (DEGs) were identified using Seurat's FindMarkers function. Next, key cell type pairs were predicted by DeepLinc, and Ligand-Receptor (LR) analysis of the DEGs in these cell types was performed using the NicheNet (Browaeys et al. 2020). Subsequently, we utilized ClusterProfiler to perform Gene Ontology (GO) enrichment and KEGG pathway analysis for the dysregulated LR-associated genes. Our analysis reveals significant disruptions in CCIs in the MTG region of AD. We identified decreased astrocyte and microglial signaling to neurons, downregulated communication between excitatory and inhibitory neurons, and upregulated microglia-to-astrocyte and oligodendrocyte interactions. Additionally, abnormal interactions between endothelial cells and other brain cells were observed. Our integrative analysis identified key dysregulated LR interactions in a cell type-specifc manner and key biological functions and pathways associated with them. This study highlights the utility of VGAEs for analyzing spatially resolved single-cell transcriptomic data to restore disrupted cellular communication in AD by unsupervised integration of both the entire transcriptomic landscape and graph topology.