A Statistical Approach of Background Removal and Spectrum Identification for SERS Data

Abstract

SERS (surface-enhanced Raman scattering) enhances the Raman signals, but the plasmonic effects are sensitive to the chemical environment and the coupling between nanoparticles, resulting in large and variable backgrounds, which make signal matching and analyte identification highly challenging. Removing background is essential, but existing methods either cannot fit the strong fluctuation of the SERS spectrum or do not consider the spectra's shape change across time. Here we present a new statistical approach named SABARSI that overcomes these difficulties by combining information from multiple spectra. Further, after efficiently removing the background, we have developed the first automatic method, as a part of SABARSI, for detecting signals of molecules and matching signals corresponding to identical molecules. The superior efficiency and reproducibility of SABARSI are shown on two types of experimental datasets.

Figure 1

Work flow of SABARSI. The phrases in the boxes show the data used or obtained, and the phrases on the side of arrows show the operations and the algorithms used.

Figure 2

Performance of four different BCMs on the spectrum of riboflavin. (a) The original spectra (black lines) and the estimated backgrounds (red lines) by NMM, PF, IRLS, and SABARSI. (b) the corresponding background-corrected spectra (black lines). The two blue boxes in the leftmost figures highlight the two regions where NMM performs poorly. (c) Background-corrected spectra by NMM at four different time points, 3,550, 3,560, 3,570, 3,580, and 3,590. The blue box corresponds to the second blue box in (b). Apparently, these are backgrounds that have not been successfully removed. (d) Background-corrected spectra by SABARSI at the same set of four different time points. With SABARSI, the background has been removed thoroughly, highlighting the true signals (red lines in the 650~900 frequency range).

Figure 3

Change in the shape of backgrounds generated by SABARSI and CBC. (a) Spectra at five different time points (represented by different colors). Each spectrum is scaled by its average intensity to facilitate the comparison of the shape. Zoomed-in regions of spectrum fragments marked in the blue boxes are shown in the second row. (b) From left to right, the three plots show the original unprocessed spectra of riboflavin, background-removed spectra by SABARSI, and background-removed spectra by CBC, in five technical replicates (from top to bottom).

Figure 4

Spectra and signals of riboflavin from replicate 3 (blue) and replicate 4 (red). (a) Two spectra are scaled by their average intensities to be comparable in the same intensity scale. The fragments of signals in the frequency region 500~1,000 before (b) and after (c) the optimal frequency shift. The correlation coefficient of the two lines increases from 0.025 to 0.719 after the shift.

Figure 5

Identified signals corresponding to the spiked reference molecule. (a) The average intensities of spectra at different time points in three replicates of the tumor lysate data. The largest intensity that corresponds to the spiked reference molecule appears at time point 2,919, 2,929, and 2,893, respectively. (b) The signals identified in the three replicates that correspond to the spiked reference molecule. The pairwise Pearson’s correlations for the three signals are around 0.8.