This review summarizes recent progress made using Raman spectroscopy and machine learning for potential universal medical diagnostic applications.

Surface- and tip-enhanced Raman spectroscopy (SERS and TERS) are modern spectroscopic techniques, which are becoming widely used and show a great potential for the structural characterisation of biological systems. Strong enhancement of the Raman signal through localised surface plasmon resonance enables chemical detection at the single-molecule scale. Enhanced Raman spectra collected from biological specimens, such as peptides, proteins or microorganisms, were often observed to lack the amide I band, which is commonly used as a marker for the interpretation of the secondary protein structure. The cause of this phenomenon was unclear for many decades. In this work, we investigated this phenomenon for native insulin and insulin fibrils using both TERS and SERS and compared these spectra to the spectra of well-defined homo peptides. The results indicate that the appearance of the amide I Raman band does not correlate with the protein aggregation state, but is instead determined by the size of the amino acid side chain. For short model peptides, the absence of the amide I band in TERS and SERS spectra correlates with the presence of a bulky side chain. Homo-glycine and -alanine, which are peptides with small side chain groups (H and CH₃, respectively), exhibited an intense amide I band in almost 100% of the acquired spectra. Peptides with bulky side chains, such as tyrosine and tryptophan, exhibited the amide I band in 70% and 31% of the acquired spectra, respectively.

A two-step process of protein detection at a single molecule level using SERS was developed as a proof-of-concept platform for medical diagnostics. First, a protein molecule was bound to a linker in the bulk solution and then this adduct was chemically reacted with the SERS substrate. Traut’s Reagent (TR) was used to thiolate Bovine serum albumin (BSA) in solution followed by chemical cross linking to a gold surface through a sulfhydryl group. A Glycine-TR adduct was used as a control sample to identify the protein contribution to the SER spectra. Gold SERS substrates were manufactured by electrochemical deposition. Solutions at an ultralow concentration were used for attaching the TR adducts to the SERS substrate. Samples showed the typical behavior of a single molecule SERS including spectral fluctuations, blinking and Raman signal being generated from only selected points on the substrate. The fluctuating SER spectra were examined using Principle Component Analysis. This unsupervised statistics allowed for the selecting of spectral contribution from protein moiety indicating that the method was capable of detecting a single protein molecule. Thus we have demonstrated, that the developed two-step methodology has the potential as a new platform for medical diagnostics.