Časoprostorová grafová neuronová síť pro včasnou detekci rakoviny prostřednictvím dynamiky cytokinové signalizace

Goldwein, Robert

Title: A Spatiotemporal Graph Neural Network for Early Cancer Detection via Cytokine Signaling Dynamics Author: Robert Goldwein Department: Department of Theoretical Computer Science and Mathematical Logic Supervisor of the master thesis: Mgr. Roman Neruda, CSc., KTIML Abstract: Early cancer often goes unnoticed by conventional imaging or molecular assays until later stages, limiting treatment options and efficacy. However, mounting evidence suggests that subtle, time-dependent shifts in circulating cytokine profiles precede detectable malignancy. Cytokines- small proteins mediating intercellular communication-form complex signaling networks whose dynamics may harbor predictive signatures of tumorigenesis. This thesis introduces a Spatiotemporal Graph Neural Network (STGNN) framework to model and forecast these cytokine trajectories, enabling extremely early cancer detection. Leveraging publicly available time-course data (e.g., CytoSig) and a novel hybrid architecture-combining graph convolutions with temporal recurrence and a biologically informed edge-decay term-it predicts both future cytokine states and an early-warning risk score for an imminent "cancer-like" signaling switch. Applied to colorectal cancer datasets, our framework not only forecasts cytokine dynamics with greater accuracy but also...

guest :: login Digital Repository
		Search		Submit		Help		About