Prediction of Ligand-Receptor Interactions Based on CatBoost and Deep Forest and Their Application in Cell-Cell...

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Cell-to-cell communication (CCC) is crucial for biological processes, including growth, development, and tissue formation.
• Understanding cellular interplay is key to deciphering complex diseases and identifying therapeutic targets.
• Ligand-receptor pair (LRP) interactions mediate cellular dialogue, forming the basis of CCC inference.

Purpose of the Study:

• To introduce CellCDmT, a novel computational method for elucidating cellular crosstalk.
• To accurately interpret LRP candidates and quantify LRI-mediated communication strength.
• To visualize intercellular and intracellular communication networks for disease mechanism insights.

Main Methods:

• CellCDmT vectorizes LRPs using PyFeat and selects features via XGBoost for classification.
• An ensemble model combining CatBoost and Deep Forest classifies unlabeled LRPs.
• Communication strength is quantified using a Three-point evaluation strategy with maximum difference, followed by visualization.

Main Results:

• CellCDmT demonstrated high accuracy in classifying unlabeled LRPs and decoding cellular crosstalk.
• Benchmarking against existing tools confirmed CellCDmT's superior performance across 8 evaluation metrics.
• The method successfully visualized communication networks in breast cancer, identifying potential therapeutic targets.

Conclusions:

• CellCDmT provides a robust computational framework for CCC inference and LRI analysis.
• The identified LRPs (MIF-CD74, WNT7B-FZD1, B2M-TFRC) and ligands (FGF22, B2M, RSPO4) may be critical in breast cancer.
• CellCDmT facilitates understanding disease mechanisms, promoting targeted therapy and drug design.