Near infrared spectroscopy (NIRS) and machine learning as a promising tandem for fast viral detection in serum microsamples: A preclinical proof of concept

Fast detection of viral infections is a key factor in the strategy for the prevention of epidemics expansion and follow-up. Hepatitis C is paradigmatic within viral infectious diseases and major challenges to elimination still remain. Near infrared spectroscopy (NIRS) is an inexpensive, clean, safe method for quickly detecting viral infection in transmission vectors, aiding epidemic prevention. Our objective is to evaluate the combined potential of machine learning and NIRS global molecular fingerprint (GMF) from biobank sera as an efficient method for HCV activity discrimination in serum. GMF of 151 serum biobank microsamples from hepatitis C patients were obtained with a FT-NIR spectrophotometer in reflectance mode. Multiple scatter correction, smoothing and Saviztsky-Golay second derivative were applied. Spectral analysis included Principal Component Analysis (PCA), Bootstrap and L1-penalized classification. Microsamples of 70 µl were sufficient for GMF acquisition. Bootstrap evidenced significant difference between HCV PCR positive and negative sera. PCA renders a neat discrimination between HCV PCR-positive and negative samples. PCA loadings together with L1-penalized classification allow the identification of discriminative bands. Active virus positive sera are associated to free molecular water, whereas water in solvation shells is associated to HCV negative samples. Divergences in the water matrix structure and the lipidome between HCV negative and positive sera, as well as the relevance of prooxidants and glucose metabolism are reported as potential biomarkers of viral activity. Our proof of concept demonstrates that NIRS GMF of hepatitis C patients' sera aided by machine learning allows for efficient discrimination of viral presence and simultaneous potential biomarker identification.