CellNEST reveals cell-cell relay networks using attention mechanisms on spatial transcriptomics

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Cell-cell communication is crucial in complex diseases like cancer and diabetes.
• Current transcriptomic methods struggle to detect intercellular signals accurately and at scale.
• Existing approaches often yield high false-positive rates and only identify single ligand-receptor interactions.

Purpose of the Study:

• To develop a novel computational approach for detecting complex cell-cell communication patterns.
• To overcome limitations of existing methods in transcriptomics, such as high false-positive rates and inability to detect multi-step communication.
• To provide a tool for analyzing intercellular signaling networks in various biological contexts.

Main Methods:

• Development of Cell Neural Networks on Spatial Transcriptomics (CellNEST), a novel computational model.
• Implementation of a new relay-network communication detection mechanism to identify ligand-receptor-ligand-receptor interactions.
• Application of CellNEST to spatial transcriptomics data from human lymph nodes, lung adenocarcinoma, colorectal cancer, and pancreatic cancer.

Main Results:

• CellNEST successfully detected T cell homing signals in human lymph nodes.
• The model identified aggressive cancer communication patterns in lung adenocarcinoma and colorectal cancer.
• New putative relay network communication patterns were predicted in pancreatic cancer, suggesting novel biological mechanisms.

Conclusions:

• CellNEST offers a powerful new method for deciphering complex cell-cell communication in transcriptomics.
• The model enhances the detection of intercellular signaling, including multi-step relay networks.
• CellNEST provides valuable insights into disease mechanisms and facilitates the exploration of in situ communication through interactive visualization.