Comparative network analysis in the context of single-cell transcriptome-derived cell-cell communication
RWTH Publications (RWTH Aachen)
Abstract
Graphs are mathematical structures used to model relationships between entities and are widely applied to complex real-world data, including social, geographic, and molecular systems. In computational biology, advances in single-cell RNA sequencing (scRNA-seq) have led to extensive molecular datasets, where graphs are commonly used to represent and analyze biological processes in problems such as clustering and cell–cell communication (CCC). Cell–cell communication, particularly through ligand–receptor interactions, plays a crucial role in tissue function and can be naturally modeled as directed, weighted graphs in which cells are nodes and interactions are edges. Although many tools exist to inferCCC from scRNA-seq data, comparing CCC graphs remains a challenging and an underexplored problem due to the complexity of directed and weighted graph structures, noise in real-world data, and limited interpretability of existing methods. This thesis introduces new computational frameworks for comparing and analyzing CCC graphs. The contributions include: (1) CrossTalkeR, a method for prioritizing and comparing CCC graphs using node importance measures; (2) the development of optimal transport–based distance measures, including Wasserstein and Gromov–Wasserstein distances, for comparing directed weighted graphs; and (3) scACCorDiON, an optimal transport–based clustering approach for analyzing patient-specific CCCgraphs in large cohorts. Methods are validated on real-world biological case studies on bone marrow fibrosis, myocardial infarction, COVID-19 and pancreatic cancer. These demonstrate that proposed approaches can improve interpretability, disease stratification, and biological relevance of cell-cell communication-driven mechanisms.
Authors 1
-
Affiliation as printed
RWTH Aachen