Biology subjects

Degroeve, S.

Publications and source records attributed to Degroeve, S..

15 recordsLinked to original sources

bioRxiv2026

Evaluating the use of non-linear models in data-driven rescoring of peptide-spectrum matches

In mass spectrometry (MS)-based proteomics, computational tools match acquired tandem MS spectra to peptides from a sequence database. Machine learning increasingly supports this task through peptide-spectrum match (PSM) rescoring, in which a classifier, typically a linear semi-supervised model, refines the initial matching score. However, Mokapot allows the user to choose among different machine learning algorithms of increasing complexity, from the default linear support vector machine (LSVM) to random forest and XGBoost. Here, we use an entrapment approach to assess the effect of this increasing complexity on PSM identification and the accuracy of the estimated false discovery rate (FDR). We show that, while more complex models increase the number of identified PSMs at a fixed FDR threshold, this gain reflects a bias towards random matches from the target proteome database rather than genuine identifications. Indeed, for the most complex model, the entrapment FDR reaches 6.3% instead of the estimated 1% decoy FDR. This bias thus yields overly optimistic FDR estimates, indicating that model complexity in PSM rescoring must be carefully balanced against this overfitting risk.

bioinformatics↗

bioRxiv2025

iDeepLC: chemical structure information yields improved retention time prediction of peptides with unseen modifications

Deep learning has notably advanced the field of liquid chromatography-mass spectrometry-based proteomics. Accurate prediction of peptide retention times significantly enhances our ability to match LC-MS data with the correct peptides and proteins, especially for DIA data. While numerous models predict peptide LC retention times with high accuracy, few can accurately predict the retention times of chemically modified peptides, particularly those with modifications not encountered during model training. In our previously developed DeepLC model, accurate predictions could be made for unseen modifications by leveraging the chemical composition of (modified) residues. Here, however, we present a further enhancement of this model based on chemical structural information. The resulting model, called iDeepLC, shows overall more accurate predictions, and better generalization performance for predicting the retention time of unseen modifications than DeepLC. iDeepLC is freely available as open-source software under the Apache2 license and can be found at https://github.com/CompOmics/iDeepLC.

bioinformatics↗

bioRxiv2025

Unraveling the cis-regulatory code controlling abscisic acid-dependent gene expression in Arabidopsis using deep learning

Abscisic acid (ABA) is a key regulator of abiotic stress responses in plants. Understanding the regulation of ABA-dependent gene expression is key to uncovering how plants adapt to environmental stress and how their resilience can be improved. We explored gene expression regulation by ABA in Arabidopsis thaliana through the training of an interpretable deep learning model predicting ABA responsiveness in the root from proximal promoter sequences. Implementing state-of-the-art augmentation strategies to boost performance, our convolutional neural network-based model was able to confidently predict ABA responsiveness. We demonstrate that it learned actual motifs in the promoter sequences and confirm that ABRE-binding factor (ABF) binding sites play a key role in regulating ABA-dependent gene expression. However, also other motifs contributing to ABA-mediated gene expression regulation were identified. Our model outperforms a model trained to predict ABF binding, indicating it successfully learned the cis-regulatory code beyond the canonical ABF binding sites. Furthermore, the importance of motif clustering for regulating gene expression levels in response to ABA was unveiled by our model. Lastly, we identified genomic regions-beyond the proximal promoter-both with and without ABF binding sites, predicted by our model to drive ABA responsiveness. These genomic regions were used to generate reporter lines for experimental validation and were shown to drive the response to ABA in planta. This confirms that our model successfully inferred the regulatory code controlling ABA-dependent gene expression.

plant biology↗

bioRxiv2024

TIMS2Rescore: A DDA-PASEF optimized data-driven rescoring pipeline based on MS2Rescore

The high throughput analysis of proteins with mass spectrometry (MS) is highly valuable for understanding human biology, discovering disease biomarkers, identifying therapeutic targets, and exploring pathogen interactions. To achieve these goals, specialized proteomics subfields - such as plasma proteomics, immunopeptidomics, and metaproteomics - must tackle specific analytical challenges, such as an increased identification ambiguity compared to routine proteomics experiments. Technical advancements in MS instrumentation can counter these issues by acquiring more discerning information at higher sensitivity levels, as is exemplified by the incorporation of ion mobility and parallel accumulation - serial fragmentation (PASEF) technologies in timsTOF instruments. In addition, AI-based bioinformatics solutions can help overcome ambiguity issues by integrating more data into the identification workflow. Here, we introduce TIMS2Rescore, a data-driven rescoring workflow optimized for DDA-PASEF data from timsTOF instruments. This platform includes new timsTOF MS2PIP spectrum prediction models and IM2Deep, a new deep learning-based peptide ion mobility predictor. Furthermore, to fully streamline data throughput, TIMS2Rescore directly accepts Bruker raw mass spectrometry data, and search results from ProteoScape and many other search engines, including MS Amanda and PEAKS. We showcase TIMS2Rescore performance on plasma proteomics, immunopeptidomics (HLA class I and II), and metaproteomics data sets. TIMS2Rescore is open-source and freely available at https://github.com/compomics/tims2rescore.

bioinformatics↗

bioRxiv2022

PhosphoLingo: protein language models for phosphorylation site prediction

Protein post-translational modifications (PTMs) play an important role in numerous biological processes by significantly affecting protein structure and dynamics. Effective computational methods that provide a sequence-based prediction of PTM sites are desirable to guide functional experiments. Whereas these methods typically train neural networks on one-hot encoded amino acid sequences, protein language models carry higher-level pattern information that may improve sequence based prediction performance and hence constitute the current edge of the field. In this study, we first evaluate the training of convolutional neural networks on top of various protein language models for sequence based PTM prediction. Our results show substantial prediction accuracy improvements for various PTMs with current procedures of dataset compilation and model performance evaluation. We then used model interpretation methods to study what these advanced models actually base their learning on. Importantly for the entire field of PTM site predictors trained on proteomics-derived data, our model interpretation and transferability experiments reveal that the current approach to compile training datasets based on proteomics data leads to an artefactual protease-specific training bias that is exploited by the prediction models. This results in an overly optimistic estimation of prediction accuracy, an important caveat in the application of advanced machine learning approaches to PTM prediction based on proteomics data. We suggest a partial solution to reduce this data bias by implementing negative sample filtering, only allowing candidate PTM sites in matched peptides that are present in the experimental metadata. Availability and implementationThe prediction tool, with training and evaluation code, trained models, datasets, and predictions for various PTMs are available at https://github.com/jasperzuallaert/PhosphoLingo. Contactsven.degroeve@vib-ugent.be and nico.callewaert@vib-ugent.be Supplementary informationSupplementary materials are available at bioRxiv.

bioinformatics↗

bioRxiv2021

MS2Rescore: Data-driven rescoring dramatically boosts immunopeptide identification rates

Immunopeptidomics aims to identify Major Histocompatibility Complex-presented peptides on every cell that can be used in anti-cancer vaccine development. However, existing immunopeptidomics data analysis pipelines suffer from the non-tryptic nature of immunopeptides, complicating their identification. Previously, peak intensity predictions by MS2PIP and retention time predictions by DeepLC, have been shown to improve tryptic peptide identifications when rescoring peptide-spectrum matches with Percolator. However, as MS2PIP was tailored towards tryptic peptides, we have here retrained MS2PIP to include non-tryptic peptides. Interestingly, the new models not only greatly improve predictions for immunopeptides, but also yield further improvements for tryptic peptides. We show that the integration of new MS2PIP models, DeepLC, and Percolator in one software package, MS2Rescore, increases spectrum identification rate and unique identified peptides with 46% and 36% compared to standard Percolator rescoring at 1% FDR. Moreover, MS2Rescore also outperforms the current state-of-the-art in immunopeptide-specific identification approaches. Integration of immunopeptide MS2PIP models, DeepLC, and Percolator into MS2Rescore thus allows substantial improved identification of novel epitopes from existing immunopeptidomics workflows.

bioinformatics↗

bioRxiv2021

ionbot: a novel, innovative and sensitive machine learning approach to LC-MS/MS peptide identification

Mass spectrometry-based proteomics generates vast amounts of signal data that require computational interpretation to obtain peptide identifications. Dozens of algorithms for this task exist, but all exploit only part of the acquired data to judge a peptide-to-spectrum match (PSM), ignoring important information such as the observed retention time and fragment ion peak intensity pattern. Moreover, only few identification algorithms allow open modification searches that can substantially increase peptide identifications. We here therefore introduce ionbot, a novel open modification search engine that is the first to fully merge machine learning with peptide identification. This core innovation brings the ability to include a much larger range of experimental data into PSM scoring, and even to adapt this scoring to the specifics of the data itself. As a result, ionbot substantially increases PSM confidence for open searches, and even enables a further increase in peptide identification rate of up to 12% by also considering lower-ranked, co-eluting matches for a fragmentation spectrum. Moreover, the exclusive use of machine learning for scoring also means that any future improvements to predictive models for peptide behavior will also result in more sensitive and accurate peptide identification.

bioinformatics↗

bioRxiv2020

The Personalized Proteome: Comparing Proteogenomics and Open Variant Search Approaches for Single Amino Acid Variant Detection

Discovery of variant peptides such as single amino acid variant (SAAV) in shotgun proteomics data is essential for personalized proteomics. Both the resolution of shotgun proteomics methods and the search engines have improved dramatically, allowing for confident identification of SAAV peptides. However, it is not yet known if these methods are truly successful in accurately identifying SAAV peptides without prior genomic information in the search database. We studied this in unprecedented detail by exploiting publicly available long-read RNA seq and shotgun proteomics data from the gold standard reference cell line NA12878. Searching spectra from this cell line with the state-of-the-art open modification search engine ionbot against carefully curated search databases resulted in 96.7% false positive SAAVs and an 85% lower true positive rate than searching with peptide search databases that incorporate prior genetic information. While adding genetic variants to the search database remains indispensable for correct peptide identification, inclusion of long-read RNA sequences in the search database contributes only 0.3% new peptide identifications. These findings reveal the differences in SAAV detection that result from various approaches, providing guidance to researchers studying SAAV peptides and developers of peptide spectrum identification tools.

bioinformatics↗

bioRxiv2020

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

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. Moreover, we show that DeepLCs ability to predict retention times for any modification enables potentially incorrect identifications to be flagged in an open modification search of CD8-positive T-cell proteome data. 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.

bioinformatics↗

bioRxiv2020

Generalized calibration across LC-setups for generic prediction of small molecule retention times

MotivationAccurate prediction of liquid chromatographic retention times from small molecule structures is useful for reducing experimental measurements and for improved identification in targeted and untargeted MS. However, different experimental setups (e.g. differences in columns, gradients, solvents, or stationary phase) have given rise to a multitude of prediction models that only predict accurate retention times for a specific experimental setup. In practice this typically results in the fitting of a new predictive model for each specific type of setup, which is not only inefficient but also requires substantial prior data to be accumulated on each such setup. ResultsHere we introduce the concept of generalized calibration, which is capable of the straightforward mapping of retention time models between different experimental setups. This concept builds on the database-controlled calibration approach implemented in PredRet, and fits calibration curves on predicted retention times instead of only on observed retention times. We show that this approach results in significantly higher accuracy of elution peak prediction than is achieved by setup-specific models.

bioinformatics↗

bioRxiv2019

The future of peptide-centric Data-Independent Acquisition is predicted

Data-Independent Acquisition (DIA) generates comprehensive yet complex mass spectrometric data, which imposes the use of data-dependent acquisition (DDA) libraries for deep peptide-centric detection. We here show that DIA can be redeemed from this dependency by combining predicted fragment intensities and retention times with narrow window DIA. This eliminates variation in library building and omits stochastic sampling, finally making the DIA workflow fully deterministic. Especially for clinical proteomics, this has the potential to facilitate inter-laboratory comparison. Significance of the StudyData-independent acquisition (DIA) is quickly developing into the most comprehensive strategy to analyse a sample on a mass spectrometer. Correspondingly, a wave of data analysis strategies has followed suit, improving the yield from DIA experiments with each iteration. As a result, a worldwide wave of investments in DIA is already taking place in anticipation of clinical applications. Yet, there is considerable confusion about the most useful and efficient way to handle DIA data, given the plethora of possible approaches with little regard for compatibility and complementarity. In our manuscript, we outline the currently available peptide-centric DIA data analysis strategies in a unified graphic called the DIAmond DIAgram. This leads us to an innovative and easily adoptable approach based on predicted spectral information. Most importantly, our contribution removes what is arguably the biggest bottleneck in the field: the current need for Data Dependent Acquisition (DDA) prior to DIA analysis. Fractionation, stochastic data acquisition, processing and identification all introduce bias in the library. By generating libraries through data independent, i.e. deterministic acquisition, stochastic sampling in the DIA workflow is now fully omitted. This is a crucial step towards increased standardization. Additionally, our results demonstrate that a proteome-wide predicted spectral library can surrogate an exhaustive DDA Pan-Human library that was built based on 331 prior DDA runs.

bioinformatics↗

bioRxiv2019

Updated MS2PIP web server delivers fast and accurate MS2 peak intensity prediction for multiple fragmentation methods, instruments and labeling techniques

MS2PIP is a data-driven tool that accurately predicts peak intensities for a given peptides fragmentation mass spectrum. Since the release of the MS2PIP web server in 2015, we have brought significant updates to both the tool and the web server. Next to the original models for CID and HCD fragmentation, we have added specific models for the TripleTOF 5600+ mass spectrometer, for TMT-labeled peptides, for iTRAQ-labeled peptides and for iTRAQ-labeled phosphopeptides. Because the fragmentation pattern is heavily altered in each of these cases, these additional models greatly improve the prediction accuracy for their corresponding data types. We have also substantially reduced the computational resources required to run MS2PIP, and have completely rebuilt the web server, which now allows predictions of up to 100.000 peptide sequences in a single request. The MS2PIP web server is freely available at https://iomics.ugent.be/ms2pip/.

bioinformatics↗

bioRxiv2018

Accurate peptide fragmentation predictions allow data driven approaches to replace and improve upon proteomics search engine scoring functions

The use of post-processing tools to maximize the information gained from a proteomics search engine is widely accepted and used by the community, with the most notable example being Percolator - a semi-supervised machine learning model which learns a new scoring function for a given dataset. The usage of such tools is however bound to the search engines scoring scheme, which doesnt always make full use of the intensity information present in a spectrum. By leveraging another machine learning-based tool, MS2PIP, we aim to overcome this obstacle. MS2PIP predicts fragment ion peak intensities. We show how comparing these intensities to annotated experimental spectra by calculating direct similarity metrics rather than the more common peak counting or explained intensities summing provides enough information for a tool such as Percolator to accurately separate two classes of PSMs, recovering more information out of the data while maintaining control of statistics such as the false discovery rate.

bioinformatics↗

bioRxiv2018

Unbiased dynamic characterization of RNA-protein interactions by OOPS

Current methods for the identification of RNA-protein interactions require a quantity and quality of sample that hinders their application, especially for dynamic biological systems or when sample material is limiting. Here, we present a new approach to enrich RNA-Binding Proteins (RBPs): Orthogonal Organic Phase Separation (OOPS), which is compatible with downstream proteomics and RNA sequencing. OOPS enables recovery of RBPs and free protein, or protein-bound RNA and free RNA, from a single sample in an unbiased manner. By applying OOPS to human cell lines, we extract the majority of known RBPs, and importantly identify additional novel RBPs, including those from previously under-represented cellular compartments. The high yield and unbiased nature of OOPS facilitates its application in both dynamic and inaccessible systems. Thus, we have identified changes in RNA-protein interactions in mammalian cells following nocodazole cell-cycle arrest, and defined the first bacterial RNA-interactome. Overall, OOPS provides an easy-to-use and flexible technique that opens new opportunities to characterize RNA-protein interactions and explore their dynamic behaviour.

cell biology↗

bioRxiv2018

Comprehensive and empirical evaluation of machine learning algorithms for LC retention time prediction

Liquid chromatography is a core component of almost all mass spectrometric analyses of (bio)molecules. Because of the high-throughput nature of mass spectrometric analyses, the interpretation of these chromatographic data increasingly relies on informatics solutions that attempt to predict an analytes retention time. The key components of such predictive algorithms are the features these are supplies with, and the actual machine learning algorithm used to fit the model parameters.\n\nWe here therefore evaluate the performance of seven machine learning algorithms on 36 distinct metabolomics data sets, using two distinct feature sets. Interestingly, the results show that no single learning algorithm performs optimally for all data sets, with different algorithm types achieving top performance for different types of analytes or different protocols. Our results can thus be used to find an optimal retention time prediction algorithm for specific analytes or protocols. Importantly, however, our results also show that blending different types of models together decreases the error on outliers, indicating that the combination of several approaches holds substantial promise for the development of more generic, high-performing algorithms.

bioinformatics↗