StellarPath: Hierarchical-vertical multi-omics classifier synergizes stable markers and interpretable similarity networks for patient profiling

• DOI: 10.1371/journal.pcbi.1012022
• Publication Date: 2024-04-12T17:21:33Z
• Authors: Luca Giudice; Ahmed Mohamed; Tarja Malm
• Keywords: Profiling (computer programming); Similarity measure; Similarity (geometry)

The Patient Similarity Network paradigm implies modeling the similarity between patients based on specific data. can summarize patients’ relationships from high-dimensional data, such as biological omics. end PSN undergo un/supervised learning tasks while being strongly interpretable, tailored for precision medicine, and ready to be analyzed with graph-theory methods. However, these benefits are not guaranteed depend granularity of summarized clarity measure, complexity network’s topology, implemented methods analysis. To date, no patient classifier fully leverages paradigm’s inherent benefits. PSNs remain complex, unexploited, meaningless. We present StellarPath, a hierarchical-vertical that pathway analysis concepts find meaningful features both classes individuals. StellarPath processes omics hierarchically integrates them into pathways, uses novel measure how activity is alike. It selects biologically relevant molecules, networks, considering molecule stability topology. A graph convolutional neural network then predicts unknown known cases. excels in classification performances computational resources across sixteen datasets. demonstrates proficiency inferring class new described external independent studies, following its initial training testing phases local dataset. advances provides markers, insights, tools in-depth profiling.