Enhancing Vibrational Spectroscopy-Based Diagnosis through Bottom-Up Modeling: The Case of Infrared Absorption Spectrum of Urine

Vibrational spectroscopy has become a valuable tool for biomedical applications. However, the technique generates large empirical spectral data sets, requiring multivariate analysis using an ever-evolving arsenal of machine learning approaches. Optimizing and validating these algorithms requires high-quality, appropriately representative data sets with a well-defined diagnostic profile, which are not readily available for testing. This work introduces an in silico framework for building and validating spectroscopic diagnostic models using mixture-based spectral simulations, illustrated through infrared urine analysis for diagnosing diabetic kidney disease (DKD). We adopt a bottom-up approach, using a priori biological information related to the disease under investigation to construct a synthetic spectral model of liquid urine, enabling the prediction of diagnostic performance using only literature-derived parameters. Experimental variables, including the protein preconcentration device, measurement time, and instrument settings, were virtually simulated and optimized, generating large and diverse data sets that capture representative ranges of experimental and patient conditions. The resulting data sets were then used to construct machine learning models for predicting albumin and creatinine concentrations. These models obtained with the bottom-up methodology were experimentally validated by comparing with models built with artificial (N = 49) and real urine samples (N = 61, DKD case-control study), obtaining no significant differences in the analytical figures of merit. The proposed approach eliminates the need for resource-intensive empirical data sets and enables systematic performance prediction and exploration of critical parameters, facilitating rapid, application-specific optimization of vibrational spectroscopy workflows capable of generating over 200 spectra in less than 1 s on a standard desktop computer.