Integration of Infrared Molecular Fingerprinting Data in a Longitudinal Health Profiling Cohort

Abstract

This study explores the characteristics and interpretation of infrared molecular fingerprints (IMFs)—blood-based profiles that capture broad molecular information—for applications in precision medicine. Using data from 4,196 healthy individuals across five longitudinal visits, we integrated Fourier-transform infrared (FTIR) spectroscopy with routine clinical chemistry tests. IMF measurements showed high inter-individual and low intra-individual variability, indicating stable and unique molecular profiles over time. Machine learning models re-identified individuals with over 90% accuracy based solely on their IMF data, highlighting strong individual specificity. To enhance diagnostic resolution, we quantified within-person and between-person variability across the infrared spectrum. A tree-based optimization algorithm stratified individuals into sub-cohorts by maximizing the Index of Individuality, minimizing between-person variability to levels close to within-person. The algorithm was based on 27 blood parameters and three demographic variables, producing hierarchical splits based on averaged longitudinal values. We further modeled the relationships between IMFs and clinical parameters using linear regression, revealing robust, biologically interpretable associations. To uncover latent physiological structure, we applied Pareto Task Inference (ParTI), which identified a tetrahedral organization in the combined IMF-clinical data space, representing four archetypal physiological states. Individual trajectories within this space may serve as early indicators of health deviation. Archetypes were further characterized using demographic and health-related data, supporting hypotheses on systemic trade-offs in health maintenance.