PriorCCI: Interpretable Deep Learning Framework for Identifying Key Ligand-Receptor Interactions Between Specific...

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Understanding cell-cell interactions (CCI) is crucial for studying biological processes like immune responses and cancer.
• Current methods using single-cell RNA sequencing (scRNA-seq) data struggle with sparsity and heterogeneity, often missing key interactions.
• Existing statistical and network-based techniques lack the precision to prioritize biologically significant CCI.

Purpose of the Study:

• To develop a deep learning framework, PriorCCI, for scalable, interpretable, and biologically meaningful identification of CCI from scRNA-seq data.
• To overcome limitations in current methods for prioritizing ligand-receptor interactions in complex biological systems.
• To provide a robust method for analyzing gene-gene interactions across cell types, particularly within tumor microenvironments.

Main Methods:

• Developed PriorCCI, a deep learning framework utilizing a convolutional neural network (CNN).
• Integrated Grad-CAM++, an explainable artificial intelligence algorithm, for visual interpretation of gene-pair contributions.
• Applied the framework to single-cell RNA sequencing data from complex environments, such as tumors.

Main Results:

• PriorCCI effectively prioritizes interactions between cancer cells and other cell types in the tumor microenvironment.
• The framework accurately identifies biologically significant interactions, including those related to angiogenesis.
• Visual interpretation of gene-pair contributions enhances the robustness of inferred gene-gene interactions.

Conclusions:

• PriorCCI offers a powerful and interpretable approach for systematic identification and prioritization of CCI from scRNA-seq data.
• The framework addresses challenges posed by data sparsity and heterogeneity in complex biological samples.
• PriorCCI facilitates deeper insights into cellular communication networks within physiological and pathological contexts.