A Comprehensive Evaluation of MS/MS Spectrum Prediction Tools for Shotgun Proteomics

Abstract

Spectrum prediction using machine learning or deep learning models is an emerging method in computational proteomics. Several deep learning-based MS/MS spectrum prediction tools have been developed and showed their potentials not only for increasing the sensitivity and accuracy of data-dependent acquisition search engines, but also for building spectral libraries for data-independent acquisition analysis. Different tools with their unique algorithms and implementations may result in different performances. Hence, it is necessary to systematically evaluate these tools to find out their preferences and intrinsic differences. In this study, multiple datasets with different collision energies, enzymes, instruments, and species, are used to evaluate the performances of the deep learning-based MS/MS spectrum prediction tools, as well as, the machine learning-based tool MS2PIP. The evaluations may provide helpful insights and guidelines of spectrum prediction tools for the corresponding researchers.

Deep learning boosts sensitivity of mass spectrometry-based immunopeptidomics

“…Artificial intelligence is revolutionizing many areas of research and is also making substantial contributions to mass spectrometry-based proteomics. As we and others have shown recently 7 , 27 , it is possible to predict the retention times and tandem mass spectra of tryptic peptides with great precision which has led to substantial improvements in terms of the comprehensiveness and quality of proteomic experiments. Here, we demonstrated that the prediction of spectra using the deep learning architecture Prosit can be generalized to the simultaneous analysis of tryptic and non-tryptic peptides which greatly extends the range of experiments to which such predictions can be applied.…”

Artificial intelligence for proteomics and biomarker discovery

“…MS2PIP is an ensemble learning (random forest and XGBoost)-based tool, which considers chemical properties of amino acids and other fragment-related features for spectrum prediction (Degroeve et al, 2013). Recently, MS2PIP achieved 0.9 to 0.95 median Pearson correlation coefficients (R) between experimental and predicted spectra on different test datasets (Xu et al, 2020). Mobile proton model-based tools such as MassAnalyzer (Zhang, 2004) and MS-simulator (Sun et al, 2012) incorporate several assumptions to simplify this problem and, hence, made their prediction interpretable.…”

Deep Learning in Proteomics

“…Comprehensive independent benchmarking of existing tools is essential to guide the selection of methods for real applications. Recently, Xu et al [80] benchmarked three deep learning-based tools (Prosit, pDeep2 and Guan's work 46 ) and one traditional machine learning-based tool (MS 2 PIP) for MS/MS spectra prediction. The results showed that the deep learning-based tools outperform MS 2 PIP and the performance of deep learning-based tools may vary across different datasets and different peptide precursor charge states.…”