PepPre: Promote Peptide Identification Using Accurate and Comprehensive Precursors.

Accurate and comprehensive peptide precursor ions are crucial to tandem mass-spectrometry-based peptide identification. An identification engine can derive great advantages from the search space reduction enabled by credible and detailed precursors. Furthermore, by considering multiple precursors per spectrum, both the number of identifications and the spectrum explainability can be substantially improved. Here, we introduce PepPre, which detects precursors by decomposing peaks into multiple isotope clusters using linear programming methods. The detected precursors are scored and ranked, and the high-scoring ones are used for subsequent peptide identification. PepPre is evaluated both on regular and cross-linked peptide data sets and compared with 11 methods. The experimental results show that PepPre achieves a remarkable increase of 203% in PSM and 68% in peptide identifications compared to instrument software for regular peptides and 99% in PSM and 27% in peptide pair identifications for cross-linked peptides, surpassing the performance of all other evaluated methods. In addition to the increased identification numbers, further credibility evaluations evidence the reliability of the identified results. Moreover, by widening the isolation window of data acquisition from 2 to 8 Th, with PepPre, an engine is able to identify at least 64% more PSMs, thereby demonstrating the potential advantages of wide-window data acquisition. PepPre is open-source and available at http://peppre.ctarn.io.

Comparative Analysis of Chemical Cross-Linking Mass Spectrometry Data Indicates That Protein STY Residues Rarely React with N-Hydroxysuccinimide Ester Cross-Linkers.

When it comes to mass spectrometry data analysis for identification of peptide pairs linked by N-hydroxysuccinimide (NHS) ester cross-linkers, search engines bifurcate in their setting of cross-linkable sites. Some restrict NHS ester cross-linkable sites to lysine (K) and protein N-terminus, referred to as K only for short, whereas others additionally include serine (S), threonine (T), and tyrosine (Y) by default. Here, by setting amino acids with chemically inert side chains such as glycine (G), valine (V), and leucine (L) as cross-linkable sites, which serves as a negative control, we show that software-identified STY-cross-links are only as reliable as GVL-cross-links. This is true across different NHS ester cross-linkers including DSS, DSSO, and DSBU, and across different search engines including MeroX, xiSearch, and pLink. Using a published data set originated from synthetic peptides, we demonstrate that STY-cross-links indeed have a high false discovery rate. Further analysis revealed that depending on the data and the search engine used to analyze the data, up to 65% of the STY-cross-links identified are actually K-K cross-links of the same peptide pairs, up to 61% are actually K-mono-links, and the rest tend to contain short peptides at high risk of false identification.

pDeep3: Toward More Accurate Spectrum Prediction with Fast Few-Shot Learning.

Spectrum prediction using deep learning has attracted a lot of attention in recent years. Although existing deep learning methods have dramatically increased the prediction accuracy, there is still considerable space for improvement, which is presently limited by the difference of fragmentation types or instrument settings. In this work, we use the few-shot learning method to fit the data online to make up for the shortcoming. The method is evaluated using ten data sets, where the instruments includes Velos, QE, Lumos, and Sciex, with collision energies being differently set. Experimental results show that few-shot learning can achieve higher prediction accuracy with almost negligible computing resources. For example, on the data set from a untrained instrument Sciex-6600, within about 10 s, the prediction accuracy is increased from 69.7% to 86.4%; on the CID (collision-induced dissociation) data set, the prediction accuracy of the model trained by HCD (higher energy collision dissociation) spectra is increased from 48.0% to 83.9%. It is also shown that, the method is not critical to data quality and is sufficiently efficient to fill the accuracy gap. The source code of pDeep3 is available at http://pfind.ict.ac.cn/software/pdeep3.