Neural Network Insights from the SABPA study: GDF-15 as a Central Node Linking Neurophysiological Stress and Metabolic Pathways in Human Biomarker Clusters

Background: Growth Differentiation Factor 15 (GDF-15) is increasingly recognized as a dynamic systemic biomarker indicative of cellular stress states across neuroendocrine, metabolic and inflammatory pathways. Despite growing evidence, the extent to which GDF-15 reflects coordinated biomarker networks encompassing neuroendocrine and metabolic axes remains to be fully elucidated in general populations. The objective of this exploratory, hypothesis generating study was to apply multilayer perceptron neural network (MLP-NN) methodology to identify and characterize the stress-related biomarker constellation most closely associated with GDF-15 in a well-defined epidemiological cohort to understand GDF-15’s role as a systems-level stress marker. Methods: We analyzed biomarker data from a South African prospective cohort study (N=378; 47% Female; age: 46±12 years) measuring baseline levels of GDF-15 (Olink CVDIII panel) alongside a panel of glial activation and neurotrophy (S100B), catecholaminergic (urinary norepinephrine), neuroendocrine (serum ACTH, DHEA) and metabolic (plasma tyrosine, phenylalanine) markers. An MLP-NN model was physiologically informed, trained and tested to detect clustering patterns and associative weight networks linked to GDF-15. Model performance was validated using cross-validation and network consistency metrics to ensure robustness of biomarker associations. Results: MLP-NN consistently revealed a stable biomarker network clustered around GDF-15, with S100B, plasma tyrosine, phenylalanine, urinary norepinephrine, and serum ACTH and DHEA forming tightly correlated nodes (all synaptic weighting 90% and above, all P< 0.001). The MLP neural network successfully identified a stable, high-dimensional biomarker cluster with GDF-15 as input layer (Model Performance R2=0.71, error =0.18), independent of traditional risk factors (age, sex, ethnicity, energy expenditure, drinking, smoking, glucose, insulin, medication usage, cholesterol (Total:HDL)). Feature importance and connectivity analysis demonstrated tightly coupled relationships between GDF-15 and S100B (Normalized importance (NI) 0.44), urinary norepinephrine (NI=0.33), plasma tyrosine (NI=0.27) and DHEA (NI=0.24) as top contributory markers in the output layer, all P< 0.001. These associations were highly statistically significant and consistently reproduced across training and validation sets, suggesting a non-linear, stable regulatory network. Conclusions: The application of MLP-NN methodology effectively uncovered a novel integrative stress signature, identifying GDF-15 as a centerpiece link as a multidimensional neuro-endocrine-metabolic biomarker. Therefore, GDF-15 may act as both a marker of systemic energetic stress, and central node in organismal stress adaptation. Finally, this data-driven analytic approach advances understanding of GDF-15’s systemic functions and underscores the multifunctional role of GDF-15, reflecting energetic stress as well as orchestrating adaptation to complex physiological challenges. This abstract was presented at the American Physiology Summit 2026 and is only available in HTML format. There is no downloadable file or PDF version. The Physiology editorial board was not involved in the peer review process.