Related Experiment Video
Updated: Jan 8, 2026
MAME Models for 4D Live-cell Imaging of Tumor: Microenvironment Interactions that Impact Malignant Progression
Published on: February 17, 2012
DeepPNCC: reconstructing pseudo-spatial cell-cell interaction landscapes from single-cell data to decipher breast
Xu-Hua Li1,2,3, Xiao-Ling Gao4, Deng-Hui Guo1
1School of Intelligent Medicine and Technology (Big Data Research Center), Hainan Medical University, Haikou, PR China.
Background:
Single-cell RNA sequencing (scRNA-seq) enables high-resolution transcriptomic analysis but loses spatial context due to tissue dissociation, thereby limiting insights into spatially regulated cell-cell interactions critical for tissue function and disease, including cancer. Consequently, developing approaches capable of extracting spatially relevant intercellular interaction information from scRNA-seq data is of substantial importance for elucidating mechanisms underlying disease initiation and progression.
Methods:
Here, we present DeepPNCC, a novel deep learning framework that reconstructs pseudo-spatial cell-cell interaction networks (PSCCIs) from scRNA-seq data by leveraging latent spatial cues preserved in undissociated cell aggregates. Built on a variational graph autoencoder (VGAE) enhanced with adversarial regularization, DeepPNCC uniquely integrates local adjacency matrices derived from multiplet data, without relying on prior knowledge such as ligand-receptor pairs, to infer global, spatially informed interaction landscapes. DeepPNCC is available as an open-source Python package.
Results:
DeepPNCC outperforms existing methods in recovering interactions aligned with spatial transcriptomics across mouse brain and breast cancer datasets. In breast cancer, the inferred cell-cell interaction network reveals a closed-loop signaling axis in triple-negative breast cancer, in which cancer-associated fibroblasts (CAFs) and perivascular-like cells (PVL) promote angiogenesis through the ADM-CALCRL-STAT1-CD40/CCL2/ICAM1 pathway. These findings provide important clues for elucidating the regulatory mechanisms of the tumor microenvironment.
Conclusions:
DeepPNCC provides an efficient tool for spatially informed analysis of scRNA-seq data, enabling the reconstruction of pseudo-spatial cell-cell interaction networks from datasets lacking explicit spatial information. This approach expands the spatial analytical scope of scRNA-seq and facilitates mechanistic dissection of cellular ecosystems in both health and disease.
