Fast Deisotoping Algorithm and Its Implementation in the MSFragger Search Engine

Abstract

Deisotoping, or the process of removing peaks in a mass spectrum resulting from the incorporation of naturally occurring heavy isotopes, has long been used to reduce complexity and improve the effectiveness of spectral annotation methods in proteomics. We have previously described MSFragger, an ultrafast search engine for proteomics, that did not utilize deisotoping in processing input spectra. Here, we present a new, high-speed parallelized deisotoping algorithm, based on elements of several existing methods, that we have incorporated into the MSFragger search engine. Applying deisotoping with MSFragger reveals substantial improvements to database search speed and performance, particularly for complex methods like open or nonspecific searches. Finally, we evaluate our deisotoping method on data from several instrument types and vendors, revealing a wide range in performance and offering an updated perspective on deisotoping in the modern proteomics environment.

Keywords: MSFragger; deisotoping; nonspecific search; open search; preprocessing; proteomics; spectrum processing.

Figure 1.

Left: An overview of the algorithm. Right: Schematic representation of the deisotoping algorithm. The original raw spectrum is denoised by removing low intensity peaks. Then a list of possible peak clusters is generated based solely on the m/z distances. This list of clusters is then filtered down to a list of high confidence clusters by comparison with an averagine distribution and categorized by monoisotopic peak. High confidence clusters are then used to decharge fragments. Fragments with no associated clusters are decharged based on parameter set by the user.

Figure 2.

Change in the number of PSMs obtained after 1% FDR filtering (blue) and search time (green) for various search methods with deisotoping relative to the same search without deisotoping. Blue bars show relative increase in PSMs obtained from closed, open, mass offset glycopeptide, and nonspecific neuropeptidome searches from left to right. Green bars show decrease in search time with deisotoping for the same searches.

Figure 3.

Change in number of PSMs obtained after 1% FDR filtering and search time for open searches on data from various instruments with deisotoping relative to the same searches without deisotoping. Blue bars show relative increase in PSMs obtained from searches of Orbitrap Q Exactive (Chick et al.), Orbitrap Fusion Lumos (Espadas et al.), timsTOF (Meier et al.), and Sciex 6600+ (Ammar et al.) data from left to right. Green bars show decrease in search time required with deisotoping for the same searches.