Differentiating Intrinsic SERS Spectra from a Mixture by Sampling Induced Composition Gradient and Independent Component Analysis Justin L. Abell, Joonsang Lee, Qun Zhao, Harold Szu, and Yiping Zhao Supplementary Material Various mixtures of BPE and MPh were prepared and 4 µl of each mixture was applied to a separate substrate. After the solvent had dried multiple points (n = 15) were measured on each substrate using the same measurement conditions previously outlined in this report. The average spectra for each mixture-treated substrate is determined and used to ascertain the indivdual BPE and MPh component spectra y 1 and y 2 , respectively. Several different reference samples were used, BPE:MPh = 100:0, 50:50, or 0:100, as x 1 . Each reference was then compared to each of the 10 other mixtures (x 2 ) to generate y 1 and y 2 . The estimated y 1 and y 2 spectra were then compared to the s 1 and s 2 (i.e. 100:0 and 0:100 BPE:MPh, respectively) and representative results are demonstrated in Fig. S2A and S2B for BPE and MPh, respectively. For most of the estimated spectra we can see a very high degree of similarity between the estimated y 1 spectra and the pure analyte signal. The high degree of cross correlation r obtained for y 1 and y 2 with s 1 and s 2 , respectively, quantitatively demonstrate very robust and accurate separation of the component signals. A plot of the r values as a function of BPE:MPh using three different reference samples is shown in Fig. S3. These plots show that the pure BPE and pure MPh are the best references to use to obtain the most accurate BPE and MPh component signals, respectively. Incidentally, using these pure samples as references yields the worst component spectra when determining the source signal of the opposite analyte when it is present at lower concentrations. Using the 50:50 BPE:MPh sample as a reference appears to work sufficiently well for both analytes at all concentrations. Because the weighting coefficient a ij represents a (relative) quantitative measure of the component signal, we compared this value to the measured intensity of the source signal x i . Figure S4A shows the normalized ratios of the calculated mixing coefficient. For each mixture, the BPE weighting coefficient a 21 was calculated and then divided by the reference weighting coefficient a 11 . Just as with the spatial mapping data in Fig. 3, all a 21 /a 11 ratios for a given reference were divided by the maximum Page 14 of 20 Analyst Electronic Supplementary Material (ESI) for Analyst This journal is © The Royal Society of Chemistry 2011