Combine multiple mass spectral similarity measures for compound identification in GC-MS

Abstract

In this study, total seven similarity measures were combined to improve the identification performance. To test the developed system, 28,234 mass spectra from the NIST replicate library were randomly split into the training set and test set. PSO algorithm was used to find the optimized weights of seven similarity measures based on the training set, and then the optimized weights were applied into the test set. Simulation study indicates that the combination of multiple similarity measures achieves a better performance than single best measure, with the identification accuracy improved by 2.2 % and 1.7% for training and test set respectively. © 2014 Springer International Publishing Switzerland.