Comparison of spectral similarity measures for compound identification

Linear and nonlinear error models were considered for compound identification using Gaussian noise. As for linear error models, additive Gaussian models were constructed and multiplicative Gaussian models were used for nonlinear error models. For each error model, three similarity measures, cosine correlation, Pearson's correlation, and Spearman's rank correlation, were implemented to compare their performance of compound identification. Furthermore, the effect of zero intensities was investigated by calculating the correlation using two schemes, OR-zero and ALL-zero methods. The simulation studies showed that the rank-based correlation, Spearman's correlation, was more robust than other correlations to all the noises and ALL-zero method provides more information than OR-zero for compound identification.