DeepPNCC: reconstructing pseudo-spatial cell-cell interaction landscapes from single-cell data to decipher breast...

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution transcriptomic data but loses spatial context.
• Understanding spatially regulated cell-cell interactions is crucial for tissue function and disease, especially cancer.
• Existing methods struggle to recover spatial information from scRNA-seq data.

Purpose of the Study:

• To develop a novel deep learning framework, DeepPNCC, for reconstructing pseudo-spatial cell-cell interaction networks (PSCCIs) from scRNA-seq data.
• To leverage latent spatial cues from undissociated cell aggregates to infer global, spatially informed interaction landscapes.
• To provide an open-source tool for spatially informed analysis of scRNA-seq data.

Main Methods:

• DeepPNCC utilizes a variational graph autoencoder (VGAE) with adversarial regularization.
• It integrates local adjacency matrices from multiplet data to infer interactions.
• The framework does not require prior knowledge of ligand-receptor pairs.

Main Results:

• DeepPNCC accurately recovers interactions aligned with spatial transcriptomics in mouse brain and breast cancer datasets.
• It identified a closed-loop signaling axis in triple-negative breast cancer involving fibroblasts and perivascular-like cells promoting angiogenesis.
• The study elucidated a specific pathway (ADM-CALCRL-STAT1-CD40/CCL2/ICAM1) critical for tumor microenvironment regulation.

Conclusions:

• DeepPNCC offers an efficient tool for reconstructing PSCCIs from scRNA-seq data lacking explicit spatial information.
• This approach expands the spatial analytical capabilities of scRNA-seq.
• It facilitates mechanistic dissection of cellular ecosystems in health and disease.