The nature of statistical learning theory
The nature of statistical learning theory
Fast training of support vector machines using sequential minimal optimization
Advances in kernel methods
Properties of Support Vector Machines
Properties of Support Vector Machines
Peptide Charge State Determination for Low-Resolution Tandem Mass Spectra
CSB '05 Proceedings of the 2005 IEEE Computational Systems Bioinformatics Conference
Automatic Quality Assessment of Peptide Tandem Mass Spectra
Bioinformatics
Improved support vector machine generalization using normalized input space
AI'06 Proceedings of the 19th Australian joint conference on Artificial Intelligence: advances in Artificial Intelligence
Classification in a normalized feature space using support vector machines
IEEE Transactions on Neural Networks
IEEE Transactions on Information Technology in Biomedicine - Special section on new and emerging technologies in bioinformatics and bioengineering
Hi-index | 0.00 |
A single mass spectrometry experiment could produce hundreds of thousands of tandemmass spectra. Several search engines have been developed to interpret tandem mass spectra. All search engines need to determine the masses of peptide ions from theirmass/charge ratios.Unfortunately,mass spectrometers do not detect the charges of ions. A current strategy is to search candidate peptides multiple times, once for each possible charge state (typically +2 or +3). However, this strategy not only wastes the search time, but also increases the risk of false positive peptide identification. This paper aims at discriminating doubly charged spectra from triply charged ones. Twenty-eight features are introduced to describe the discriminant characteristics of doubly charged and triply charged spectra. The support vector machine (SVM) technique is used to train the classifier on these 28 features. To verify the proposed method, computational experiments are conducted on two types of datasets: ISB dataset generated from the low-resolution ion-trap instrument and TOV dataset generated from the high-resolution quadrupole-time-of-flight instrument. For each type of dataset, the SVM-based classifiers are trained and tested on 20 randomly sampled subdatasets. The results show that the proposed method reaches average correct rates of 95% and 93% to discriminate doubly charged spectra from triply charged ones for the low-resolution ISB dataset and the high-resolution TOV dataset, respectively.