A Linear Predictor Based on FTIR Spectral Biomarkers Improves Disease Diagnosis Classification: An Application to Multiple Sclerosis

Abstract

Multiple sclerosis (MS) is a neurodegenerative disease of the central nervous system that can lead to long-term disability. The diagnosis of MS is not simple and requires many instrumental and clinical tests. Sampling easily collected biofluids using spectroscopic approaches is becoming of increasing interest in the medical field to integrate and improve diagnostic procedures. Here we present a statistical approach where we combine a number of spectral biomarkers derived from the ATR-FTIR spectra of blood plasma samples of healthy control subjects and MS patients, to obtain a linear predictor useful for discriminating between the two groups of individuals. This predictor provides a simple tool in which the contribution of different molecular components is summarized and, as a result, the sensitivity (80%) and specificity (93%) of the identification are significantly improved compared to those obtained with typical classification algorithms. The strategy proposed can be very helpful when applied to the diagnosis of diseases whose presence is reflected in a minimal way in the analyzed biofluids (blood and its derivatives), as it is for MS as well as for other neurological disorders.

Keywords: ATR-FTIR; discrimination; logistic regression; multiple sclerosis; multivariate analysis; plasma.

Figure 1

ATR-FTIR spectra of plasma of HC and MS patients in the high region (left panels) and fingerprint region (right panels). Pre-processing spectral analysis includes cut, rubberband baseline subtraction, and vector normalization. The stretching vibration of specific functional groups and peaks of interest are also shown (see text for details).

Figure 2

Box-plot of the spectral parameters derived from the analysis of the ATR-FTIR absorption spectra, illustrating the distribution of their values in the HC and MS groups. The p-value is indicated as * (p ˂ 0.05); ** (p ˂ 0.01); *** (p ˂ 0.001).

Figure 3

Distribution of the predicted probability for HC (upper panel) and MS (bottom panel) individuals. The dashed vertical line indicates the 0.5 score threshold. Score values ≤0.5 indicate subjects classified as healthy, score values ˃0.5 indicate subjects classified as diseased.

Figure 4

Predictive power of the obtained model in distinguishing between HC and MS, as described by a smoothed ROC curve.