DeepLC can predict retention times for peptides that carry as-yet unseen modifications

Abstract

The inclusion of peptide retention time prediction promises to remove peptide identification ambiguity in complex LC-MS identification workflows. However, due to the way peptides are encoded in current prediction models, accurate retention times cannot be predicted for modified peptides. This is especially problematic for fledgling open modification searches, which will benefit from accurate retention time prediction for modified peptides to reduce identification ambiguity.We here therefore present DeepLC, a novel deep learning peptide retention time predictor utilizing a new peptide encoding based on atomic composition that allows the retention time of (previously unseen) modified peptides to be predicted accurately. We show that DeepLC performs similarly to current state-of-the-art approaches for unmodified peptides, and, more importantly, accurately predicts retention times for modifications not seen during training. DeepLC is available under the permissive Apache 2.0 open source license and comes with a user-friendly graphical user interface, as well as a Python package on PyPI, Bioconda, and BioContainers for effortless workflow integration.

Misincorporation Proteomics Technologies: A Review

“…Acquiring data in a DIA manner will allow retrospective interrogation for these peptides, with a caveat that the amount of NPAA-containing peptides present for analysis must be above the lower limits of detection. The ways to overcome the lack of theoretical data available to predict the effect of NPAA incorporations on produced MS/MS spectra is through the use of tools (such as DeepLC [ 82 ]) that enable the prediction of retention times and elution windows for these unknown species. This allows smarter scheduling and the ability to potentially target tens of thousands of precursors in a single injection.…”

Deep Learning in Proteomics

“…DeepLC showed comparable performance to the state-of-the-art RT prediction algorithms for unmodified peptides and achieved similar performance for unseen modified peptides to that for the unmodified peptides. [47] Other RT prediction models, such as DeepDIA and a model we developed called AutoRT, combine both CNN and RNN in the same networks. In DeepDIA, one-hot encoded peptide sequences are fed into a CNN network, which is followed by a BiLSTM network.…”

“…Machine learning‐based methods can be further divided into two sub groups: traditional machine learning‐based methods including Elude [ 42,43 ] and GPTime, [ 38 ] and deep learning‐based methods including DeepRT, [ 44 ] Prosit, [ 29 ] DeepMass, [ 45 ] Guan et al., [ 46 ] DeepDIA, [ 30 ] AutoRT, [ 25 ] and DeepLC. [ 47 ] As shown in Table 1 , deep learning‐based tools can be divided into three groups based on the type of neural network architecture used: RNN‐based, CNN‐based, and hybrid networks, with RNN as the dominant architecture because it was developed for sequential data modeling. Several of these tools also have a separate module for MS/MS spectrum prediction (see next Section).…”

“…DeepLC showed comparable performance to the state‐of‐the‐art RT prediction algorithms for unmodified peptides and achieved similar performance for unseen modified peptides to that for the unmodified peptides. [ 47 ]…”

The use of hybrid data-dependent and -independent acquisition spectral libraries empower dual-proteome profiling

“…In total, 1,784,677 MS2 spectra were predicted for 2+/3+ peptide precursors (395 to 905 m/z) for 1,295,652 peptides with enzymatic settings similar to PECAN search settings (Trypsin/P, 7 to 25 amino acid long peptides with a maximum of one missed cleavage). DIA-based peptide RTs obtained from the PECAN search results were used to train and predict RTs for all peptides using DeepLC [40]. We then used the MS²PIP-predicted spectral library to query our artificial mixtures with EncyclopeDIA ( Figure 3, left panel, grey bars).…”