Peak Selection Sample Clauses

Peak Selection. Most peak selection methods are based on simple heuristics. In general, spectral peaks are identified as the local maxima in predefined neighborhoods of intensity values. It is also common that these local maxima are required to be higher than the average intensity level of its neighborhood by a certain margin. For example, Xxx et al., (2003)[76] declare a point in the spectrum as a selected peak if the intensity is a local maximum; its absolute value is larger than a threshold; and the intensity is larger than a threshold times the average intensity in the neighborhood surrounding the peak. Xxxxxxx et al., (2003) [27] consider two peak identification procedures. In the simple peak detection procedure, local maxima are identified, and nearby maxima that likely represent the same peptides are merged after filtering out local maxima that are likely to be noise. In the simultaneous peak detection and baseline correction procedure; simple peak detection is first used to obtain a preliminary list of peaks, after which a baseline is calculated by removing these preliminary candidate peaks. A final set of peaks is selected by iterating the above steps, until all peaks selected are above a chosen signal-to-noise ratio threshold. The success of these peak selection methods depend largely on whether or not the noise model is correctly specified. Often, a particular noise model is assumed without attempting to validate the assumed model. Xxxxxxx et al., (2003)[27] define noise as the median absolute value of intensities. Xxxxxx et al., (2004)[108] use the negative part of the normalized MS data to estimate the variance of noise. Xxxxx et al., (2003a)[120] argue that binary peak/non-peak data is more useful than the absolute values of intensities in quantifying noise, based on the observation that the measured intensities are susceptible to substantial measurement error. Wavelets based approaches for de-noising have also been proposed in Coombes et al., (2004)[? ], and Xxxxxxxx and Xxxxx (2006)[99]. All peak selection algorithms are limited by the resolution of the MS data and the consequent overlapping effect of neighboring peaks. There is also the issue of false positive and false negative selections. Xxxxx et al., (2003b)[121] address this issue by adding a peak width constraint. Xxxxxxxx and Xxxxx (2004)[99]) choose a specific scale level after a wavelets based decomposition of the original MS data. In the case of high resolution data, Xx et al., (2004)[122]) prop...