MOE Key Laboratory of Bioinformatics, Center for Synthetic & Systems Biology, School of Life Sciences, Tsinghua University, Beijing, 100084, China.
Genome Biol. 2022 Jun 3;23(1):124. doi: 10.1186/s13059-022-02692-0.
Based on a deep generative model of variational graph autoencoder (VGAE), we develop a new method, DeepLinc (deep learning framework for Landscapes of Interacting Cells), for the de novo reconstruction of cell interaction networks from single-cell spatial transcriptomic data. DeepLinc demonstrates high efficiency in learning from imperfect and incomplete spatial transcriptome data, filtering false interactions, and imputing missing distal and proximal interactions. The latent representations learned by DeepLinc are also used for inferring the signature genes contributing to the cell interaction landscapes, and for reclustering the cells based on the spatially coded cell heterogeneity in complex tissues at single-cell resolution.
基于变分图自动编码器（VGAE）的深度生成模型，我们开发了一种新方法，名为 DeepLinc（用于细胞相互作用网络的深度学习框架），用于从单细胞空间转录组数据中从头重建细胞相互作用网络。DeepLinc 展示了从不完善和不完整的空间转录组数据中高效学习、过滤虚假相互作用以及推断缺失的远端和近端相互作用的能力。DeepLinc 学习到的潜在表示也可用于推断导致细胞相互作用景观的特征基因，并基于复杂组织中单细胞分辨率的空间编码细胞异质性对细胞进行重新聚类。
