Prediction of Kidney Disease Through the Serum Molecular Signature

Chronic kidney disease (CKD) is a major global health challenge, associated with increased cardiovascular risk, premature mortality, and substantial healthcare burden. As the disease often progresses asymptomatically, many patients are diagnosed only after significant and irreversible loss of kidney function, limiting the effectiveness of early therapeutic interventions.

This study investigated the potential of mid-infrared (MIR) spectroscopy, to acquire the serum molecular fingerprint, combined with machine learning for the rapid and minimally invasive assessment of CKD. A total of 54 bovine serum solutions mimicking the biochemical profiles of the five CKD stages were prepared by simultaneously varying the concentrations of creatinine, urea, and albumin. MIR spectra were acquired using a high-throughput microplate platform requiring only 5 µL of serum, a volume compatible with collection from a simple finger prick, while allowing the simultaneous analysis of up to 96 samples.

The acquired spectra were used to develop Partial Least Squares (PLS) regression models for key biochemical markers of kidney function. Excellent regression performance was obtained for serum creatinine and urea, with correlation coefficients exceeding 0.85. Furthermore, the same spectral data enabled accurate classification of CKD stages. Random Forest, Extreme Gradient Boosting, and Support Vector Machine models achieved areas under the receiver operating characteristic curve (AUC) above 0.98 when using normalized first-derivative spectra.

These results point to the capability of MIR spectroscopy to simultaneously quantify relevant renal biomarkers and accurately discriminate CKD stages from a single measurement. The proposed approach offers a rapid, low-volume, and high-throughput alternative to conventional laboratory methods, highlighting its potential for integration into future point-of-care systems for kidney disease monitoring and management.