DeltaPCA: A statistically robust method for analysing surface-enhanced Raman spectra for quantitative analyte detection

Abstract

Surface-enhanced Raman spectroscopy using functionalised nanoparticles or nano-engineered surfaces provides a highly selective and sensitive technique for detecting and quantifying analyte binding. However, practical diagnostic and sensing applications are hindered by technical variability due to hot-spotting and analyte drift, as well as variability in sample preparation and substrate uniformity. In this work, we introduce a novel joint experimental design and computational analysis procedure to minimize and/or control for these sources of error. Sample variability is minimized by preparing functionalised nanoparticles with and without analyte under the same conditions, and then recording and analysing difference spectra. To account for technical variability, multiple spectra are recorded from each sample. The key novelty of our analysis procedure is that all information about sample and technical variability is retained through to the final comparative analysis step, and we apply principal component analysis twice - once to extract “variance-minimized” spectra as principal loading vectors and again to distinguish between samples with and without the target analyte. Proof of principle experiments using thiolated aptamers to detect CoV-SARS-2 spike protein reveal that analyte binding show that analyte binding primarily shows up as a depletion of free covalent stretching bands, coupled to appearance of corresponding hydrogen-bonded stretching bands. This is quite different from previous work which largely focusses on changes in the “fingerprint” region where we find that the signal may be obscured by greater technical variability. Our computational analysis code can be freely downloaded from https://github.com/dlc62/DeltaPCA.