pDeep: Predicting MS/MS Spectra of Peptides with Deep Learning

Abstract

In tandem mass spectrometry (MS/MS)-based proteomics, search engines rely on comparison between an experimental MS/MS spectrum and the theoretical spectra of the candidate peptides. Hence, accurate prediction of the theoretical spectra of peptides appears to be particularly important. Here, we present pDeep, a deep neural network-based model for the spectrum prediction of peptides. Using the bidirectional long short-term memory (BiLSTM), pDeep can predict higher-energy collisional dissociation, electron-transfer dissociation, and electron-transfer and higher-energy collision dissociation MS/MS spectra of peptides with >0.9 median Pearson correlation coefficients. Further, we showed that intermediate layer of the neural network could reveal physicochemical properties of amino acids, for example the similarities of fragmentation behaviors between amino acids. We also showed the potential of pDeep to distinguish extremely similar peptides (peptides that contain isobaric amino acids, for example, GG = N, AG = Q, or even I = L), which were very difficult to distinguish using traditional search engines.

Deep learning the collisional cross sections of the peptide universe from a million training samples

“…We reasoned that a combination of very large and consistent data sets acquired by PASEF with state of the art deep learning methods would address both challenges. Due to their inherent flexibility and their ability to scale to large data sets, deep learning methods have proven very successful in genomics 30,31 and more recently in proteomics for the prediction of retention times and fragmentation spectra [32][33][34][35] .…”

AHLF: ad hoc learning of peptide fragmentation from mass spectra enables an interpretable detection of phosphorylated and cross-linked peptides

“…Training on simulated data or domain randomization Tobin et al (2017) is a powerful method to improve performance for models and reducing the generalization gap. The prediction of relative intensity profiles is well received in the proteomics community as it improves database search substantially Gabriels et al (2019); Zhou et al (2017); Gessulat et al (2019). This in combination with a sophisticated noise model may become very helpful for domain randomization and training models like AHLF in the near future.…”

A hybrid spectral library combining DIA-MS data and a targeted virtual library substantially deepens the proteome coverage

“…Before constructing a virtual spectral library, we tested the performance of several deep learning models to predict fragment ion intensities and retention time indices (iRT) for the 415 GPCR peptide precursors from the initial DIA spectral library. Distinct from the aforementioned wholeproteome virtual library approaches, we here used the deep neutral network-based models pDeep (Zhou et al, 2017) to predict fragment ion intensities and DeepRT (Ma et al, 2018) to predict iRT from GPCR peptide sequences (Fig. 1).…”