Overlapping Peak Separation Method of Mass Spectrometry Based on Resolution Enhancement Algorithm

Overlapping peak separation is an important pretreatment step in mass spectrometry data analysis. Because the resolution of the instrument is not high, overlapping peaks appears in the mass spectrum. The position of the characteristic peak can not be accurately determined and have influence on the identification of compounds. The intersection vertical method and the proportional distribution method use geometric method to resolve overlapping peaks which can not solve the problem of trailing peak and leading peak. Wavelet is mainly used to denoise and determine the number of peaks and the position of overlapping peaks. Under the condition of a large number of peaks overlapping, the overlapping peaks can not be accurately separated. In this paper, a new method based on resolution enhancement algorithm was proposed. Firstly, the peak resolution was improved by using a peak sharpening algorithm which weight the signal and its negative second order differential, solving the problem of tailing peak and leading peak. Then, the sharpened signal was transformed by continuous wavelet transform, the signal was denoised and the rough parameters of the spectral peak were estimated. Finally, taking these rough parameters as the input of curve fitting, the accurate fitting parameters could be obtained by nonlinear fitting. Under the condition of a large number of peaks overlapping, the overlapping peaks could be accurately separated. The results showed that the resolution was improved in the simulation of the original resolution after resolution. Compared with the wavelet transform method, the intersection vertical method and the proportional distribution method, the proposed method could realize the overlapping peak separation more accurately. The white Gaussian noise with different signal-to-noise ratio (SNR) was added to the data by using the “AWGN” function in MATLAB. The overlapping peaks with different signal-to-noise ratios were simulated. The results showed that the relative error of this proposed method was obviously smaller than that of the wavelet transform method, the intersection vertical method and the proportional distribution method.