GRAPH NEURAL NETWORKS FOR MODELING PROTEIN-PROTEIN INTERACTION NETWORKS AND CELLULAR SIGNALING PATHWAYS: A REVIEW

The revolutionary use of Graph Neural Networks (GNNs) in systems biology is explained in this review, with particular attention to the modeling of cellular signaling pathways and Protein-Protein Interaction (PPI) networks. In contrast to Euclidean deep learning architectures or conventional machine learning, GNNs use message-passing paradigms to integrate heterogeneous node attributes and strictly preserve topological dependencies, capturing both local and global features of complex biological systems. The study offers a thorough synthesis of specialized architectures, such as Temporal GNNs (T-GNNs) for encoding the dynamic, time-variant nature of signaling cascades and Heterogeneous GNNS (HGNNs) for modeling various biological entities. Critical applications are examined, showcasing GNNs' superior performance in drug target identification, protein function prediction, and the identification of disease modules to clarify disease mechanisms. Nevertheless, significant obstacles like computational scalability, the "black-box" interpretability issue, and the inherent noise and incompleteness of experimental interaction data are also thoroughly examined in the review. The authors conclude that improving model generalization and converting computational predictions into clinical contexts require integrating multi-omics data with knowledge graph embeddings and creating domain adaptation strategies.

کلیدواژه ها:

Graph Neural Networks ، Protein-Protein Interactions ، Cellular Signaling Pathways ، Systems Biology ، Heterogeneous Networks ، Deep Learning