An AI-Guided Integration of Molecular Networks and Single-Cell Transcriptomics using scNET

​

Ron Sheinin, Roded Sharan - Blavatnik School of Computer Science and AI, Tel Aviv University. 
Asaf Madi - Department of Pathology, Faculty of Medicine, Tel Aviv University​

 

The tumor microenvironment (TME) is composed of diverse cell types whose functional states and interactions shape cancer progression and therapeutic response. While single-cell RNA sequencing (scRNA-seq) has revealed this heterogeneity at high resolution, interpreting the functional significance of cell states remains a key challenge. Most analytical approaches rely solely on transcriptomic similarity, overlooking known molecular interactions that constrain gene programs.
We present scNET, an AI-based framework recently published in Nature Methods, which integrates scRNA-seq data with protein–protein interaction (PPI) networks to derive biologically informed embeddings of both cells and genes. By jointly modeling gene-gene and cell-cell relationships , scNET enables the discovery of context-specific co-regulated gene modules and condition-dependent biological pathway activation.
 

Applying scNET to a murine glioblastoma model (GL261a), we show that it substantially improves the expression accuracy of canonical marker genes and reduces zero inflation across diverse cell types. Crucially, scNET enhances downstream pathway analysis. Gene set enrichment analysis (GSEA) on scNET-reconstructed data recovered relevant KEGG pathways for major cell populations. In a tumor immunotherapy setting, scNET revealed enrichment of eight out of nine known T cell activation pathways in CD8⁺ T cells following P-selectin inhibition—signals entirely missed in the original  data.

Our framework is published at :
Sheinin, R., Sharan, R. & Madi, A. scNET: learning context-specific gene and cell embeddings by integrating single-cell gene expression data with protein–protein interactions. Nat Methods 22, 708–716 (2025).