CELLetter: leveraging large language model and dual-stream network to identify context-specific ligand-receptor...

Area of Science:

• Computational Biology
• Systems Biology
• Genomics

Background:

• Cell-to-cell communication (CCC) is vital for multicellular organisms.
• Existing computational methods often overlook intracellular signaling and rely on static ligand-receptor (L-R) databases.

Purpose of the Study:

• To develop a novel deep learning framework, CELLetter, for identifying L-R interactions and deciphering cellular signaling.
• To integrate L-R co-expression with downstream transcription factor (TF) activity for enhanced communication analysis.

Main Methods:

• Utilized ProstT5 protein language model for feature embedding.
• Employed a dual-stream architecture with a gate mechanism for feature fusion and interaction.
• Integrated L-R pairs, scRNA-seq data, and TF activity to quantify communication strength.

Main Results:

• CELLetter demonstrated superior L-R classification performance against state-of-the-art methods.
• Predicted L-R pairs showed significant spatial relevance in human heart and lung tissues.
• Identified MIF-CD44 as a key signaling axis in HNSCC, suggesting therapeutic potential.

Conclusions:

• CELLetter offers a robust framework for identifying L-R interactions and inferring CCC.
• The framework enhances understanding of cellular signaling pathways and identifies potential therapeutic targets in diseases like HNSCC.