Posters - Schedules

Posters Home

With the widespread use of MS-based shotgun proteomics, countless datasets of different human cell types and tissues with deep proteome coverage are constantly being added to repositories providing valuable quantitative information on proteome-wide protein copy numbers. However, it remains largely underused because of technical challenges to compare protein levels across individual studies. Here we introduce MaxQuantAtlas, a software platform for the integration of MaxQuant-processed proteomics datasets over many samples acquired with label-free and label-based quantification strategies and instrument types.
MaxQuantAtlas produces one unified database table with concentration profiles over all samples. For isobaric labeling samples, we introduce algorithms for ratio decompression and combined MS1-MS2 quantification to obtain cellular protein abundances. Protein-level aggregated MS signals are scaled to concentrations using a proteomic ruler-like method. For multivariate analysis, we developed a novel imputation method compatible with varying dynamic ranges. A two-dimensional quality-score based on sample dynamic ranges and correlation of housekeeping proteins detects problematic samples leading to their automatic exclusion. We clearly observed meaningful clustering of samples by biological origin that was irrespective of quantification method. Samples yielded similar concentration profiles when they were analyzed in a label-free or multiplexed together with unrelated samples of TMT sets, showing successful integration across technologies.

Understanding underlying biological processes is essential for providing effective treatment for diseases. Investigating complex biological behavior at the cellular and molecular levels requires profiling different aspects of human biology, such as small molecules (metabolites). Liquid chromatography-mass spectrometry (LC-MS) based methods are ideal tools for characterizing and investigating human health by profiling metabolites. LC-MS-based metabolomics methods can yield data on thousands of features, each characterized by its descriptors: a measured mass-to-charge ratio (m/z), chromatographic retention time (RT), and signal intensity (SI). LC-MS techniques are powerful and, at the same time, very sensitive and complex. In addition, this complexity increases when working with metabolites among different studies or LC-MS platforms. We present massSight, a computational tool to inspect and adjust trends, scale raw metabolite intensities, align peaks (annotated and unannotated) between separately acquired data sets, and remove redundancies in nontargeted LC-MS data arising from multiple ionization products of a single metabolite. The platform will come with a set of novel tools, complementary to LC-MS metabolite profiling techniques, to accurately profile features, align features across studies and platforms, perform scaling, and consolidate adducts and fragments of chemical compounds. massSight is open-source software with documentation available online at http://github.com/omicsEye/massSight.

Top-down proteomics (TDP) based on mass spectrometry (MS) has gained a lot of interest for intact proteoform analysis, and accurate quantification of proteoforms plays an essential role in such studies. Label-free quantification (LFQ) offers several advantages over labeled approaches, such as easy sample preparation and no extra cost for labeling, thus still the most common approach in proteoform quantitative studies. Due to the nature of the analyte, a large charge state range and isotope envelope per proteoform are observed in its MS data compared to conventional bottom-up proteomics; in turn, overlapping of proteoform signals (i.e., co-elution) is inevitable. This overlap complicates accurate quantification.
We present FLASHQuant for MS1-level LFQ data analysis in TDP, with automatic co-eluted proteoform resolution. FLASHQuant is evaluated with two spike-in datasets having varying ratios of spike-in analytes; a simple protein mixture level and a proteome-wide level. It showed accurate quantitation with high reproducibility among technical replicates. In both datasets, FLASHQuant showed a small variance in fold changes, and the coefficient of variation (CV) values of the consensus feature quantities among replicates are lower than 0.2. FLASHQuant achieved a fast runtime of ~2 min per ~1800 spectra and is freely accessible as platform-independent software at https://openms.de/application/flashdeconvq.

Hydrogen-Deuterium Exchange mass-spectrometry (HDX-MS) has emerged as a powerful technique to explore the conformational dynamics of proteins and protein complexes in solution. In the bottom-up approach to MS, deuterium uptake is reported at the level of peptides, which complicates interpretation and means ad-hoc approaches are used to resolve contradictions between overlapping peptides. Here we propose to leverage the overlap in peptides, the temporal component of the data and the correlation along the sequence dimension to infer residue-level uptake patterns. Our model treats HDX-MS as a multiple change-point problem - inferring at which residues HDX has changed. Fitting our model in the Bayesian non-parametric framework allows inference of the number of parameters, quantitative assessments of the confidence of differential HDX and uncertainty estimates of the temporal kinetics. We benchmark our approach against others using a three-way proteolytic digestion experiment and find that it out-performs other available methods. We illustrate our approach on a number of case-studies.

Tandem mass spectrometry (MS/MS) is a powerful tool for identifying peptides and proteins by ionizing and fragmenting them, which are then detected by a mass spectrometer, generating an array of masses-to-charge ratios corresponding to specific fragments. However, the computational analysis of MS/MS data suffers from a high rate of false discoveries and is prone to errors, highlighting the need for improved analysis methods. One essential step of computational mass-spectrometry is peptide-spectrum-matching (PSM) which is identifying the peptides based on the masses of their fragments. The problem is combinatorial in nature and heavily relies on the analytical tools for conversion of the spectrum to peptide sequence through a reference database search. These techniques still lack precision, speed, and versatility.

Deep learning methods started to gain popularity in solving various tasks in mass-spectrometry, performing on par with analytical techniques. Encouraged by these trends we build a language model learning a joint representation of peptides and spectrum. Furthermore, the model is finetuned to perform several tasks such as peptide identification, backbone peak annotation, and spectrum prediction.

Our model demonstrated excellent results in identifying peptides, comparable with analytical methods while offering speed, flexibility convenience of one model to solve several tasks.

In bottom-up proteomics, peptide identification is essential in the analysis of LC-MS/MS data. Comet is a classic software program that provides reasonable results, but its output quantity is sensitive to input parameters such as MS1 and MS2 tolerance. Our experiments have shown that even tiny differences in the MS1 or MS2 tolerance settings can cause the quantity of Comet results to fluctuate by up to 40%. To address this problem, we propose to use a protein feedback mechanism in Comet to obtain more robust and sensitive results. We have observed that, if a protein exists in the MS data, all sibling peptides should have a higher chance to appear in the same data as well. Using this information, we construct a protein score database during the peptide identification process. We adjust the score of the matched peptides, giving more weight to the peptides corresponding to the matching proteins in the database. Our results show that Comet with protein feedback (Comet-PF) has more consistent results under different MS tolerance settings than Comet and identifies 32.3% more PSMs than Comet.

Top-down proteomics (TDP) directly analyzes intact proteins and thus provides more comprehensive proteoform-level information than conventional bottom-up proteomics that relies on digested peptides. Significant advancements have been made in TDP in different aspects, progressively unlocking the potential of TDP in numerous biological and medical applications. However, reliable and reproducible data analysis still remains one of the major bottlenecks in TDP. A key prerequisite step for robust data analysis is the establishment of an objective estimation of proteoform-level false discovery rate (FDR) in proteoform identification. The most widely used FDR estimation scheme is FDR via target-decoy approach (TDA), which has primarily been established for bottom-up proteomics. We present evidence that the commonly used TDA-based FDR estimation may not work at the proteoform-level due to an overlooked factor, namely the erroneous deconvolution of precursor masses (simply precursor deconvolution error), which leads to incorrect FDR estimation. We argue that the conventional TDA-based FDR in proteoform identification is in fact protein-level FDR rather than proteoform-level FDR unless precursor deconvolution error rate is taken into account. To address this issue, we propose a formula to correct for proteoform-level FDR bias by combining TDA-based FDR and precursor deconvolution error rate.

MetaboLights is a database for Metabolomics experiments and their derived information. The database is cross-species, cross-technique and covers metabolite structures and their reference spectra as well as their biological roles, locations and concentrations. MetaboLights studies contain information across study design, sample origin, collection, processing, spectral data measurement, data processing and metabolite identification. MetaboLights compound library displays information outlining the chemical and biological nature of numerous metabolites.
Over the last few years, MetaboLights has experienced exponential growth with an ever increasing diversity in the studies submitted. The MetaboLights repository has over 1180 public Mass spectrometry studies (LC-MS, GC-MS, DI-MS, MS Imaging, and so on) with 340,000 raw and derived spectral data files across homo sapiens to arabidopsis thaliana. MetaboLights supports publications in a range of journals with a significant proportion not only in specialist metabolomics journals but also in journals such as Nature, Cell and PLOS.
MetaboLights also provides an open access Galaxy platform, MetaboLights Labs, that researchers can access and use to analyse their data with the pre-defined workflows optimised for MS pre-processing and processing tasks prior for submission to MetaboLights and additionally for the ongoing reanalysis of existing MetaboLights studies.

The potential of non-targeted metabolomics to uncover new biological insights, identify biomarkers or monitor clinical disease progression cannot be emphasized enough. However, spectral reference data is incomplete, and most compound mass spectra in non-targeted metabolomics experiments cannot be annotated with spectral search alone. At the same time, the identification and classification of unknown compounds are far from trivial. One reason is the current lack of understanding about how new molecules will fragment when subjected to tandem mass spectrometry (MS/MS). Existing in silico fragmentation methods, such as MetFrag and CFM-ID, imitate the fragmentation process but their accuracy is limited due to the way they integrate and engineer molecular features. We investigate the ability of graph neural networks (GNNs) to learn and recognize relevant structural groups associated with bond cleavage during MS/MS.

GNNs have been used with great success in the context of drug property prediction. In non-targeted metabolomics, however, graph models have received little attention and have only been used for direct property and peak predictions rather than as a foundation of in silico fragmentation frameworks. We deploy and test various types of GNNs and well-established graph models from other domains to predict the break tendencies of bonds during MS/MS fragmentation.

Nuclear proteins are major constituents and regulators of the topological organization of nucleome. To decipher the global connectivity of nuclear proteins and the hierarchically organized modules of nuclear protein-protein interactions (PPIs), the double chemical-crosslinking coupled with mass spectrometry (XL-MS) analysis was integrated with the dimethyl-labelling to generate a qXL-MS workflow, resulting in 5,340 nuclear crosslinks and 1,297 nuclear PPIs. There were 250 and 26 novel interactors of histones and nucleolar box C/D snoRNP complex, respectively. Modulomic analysis of the Arabidopsis orthoglous PPIs constructed 27 and 24 master nuclear protein-protein interaction modules (NPIMs) that contain the condensate-forming protein(s) and the intrinsically disordered region (IDR)-containing protein(s), respectively. These NPIMs successfully captured the reported nuclear protein complexes and nuclear bodies. Interestingly, these NPIMs are hierarchically assorted into 4 higher order communities in nucleomic graph, including Genome and Nucleolus Community. This combinatorial pipeline of 4C (double chemical crosslinking, chromatographic enrichment, computation analysis and confirmation) quantitative interactomics and modulomic revealed 17 hormone-specific module variants participating in a broad range of nuclear events. It is capable of capturing both nuclear protein complexes and nuclear bodies, constructing topological architectures of both PPI modules and module variants in nucleome, and potentially mapping protein compositions of biomolecular condensates.

Background

Metabolomics data analysis is a complex and multistep process, which is constantly evolving with the development of new analytical technologies, mathematical methods, bioinformatics tools and databases. By a common effort from several institutional structures as MetaboHUB (French national infrastructure) and the IFB (French ELIXIR node), Workflow4Metabolomics (W4M) endeavours to break through the barriers that are obstructing data analysis practices in this field.

Technological and methodological innovation

Workflow4Metabolomics provides free open-source Galaxy-based workflow possibilities for MS- and NMR-based data analysis as well as training contents, to help the community getting access to advanced tools and data analysis knowledge. W4M latest updates include the integration of a new tool suit for GC-MS data analysis in its supported usegalaxy.fr host, provided by RECETOX, Masaryk University (Czech EIRENE node). New training resources through the Galaxy Training Network (GTN) are also proposed, and Training Infrastructure as a Service (TIass) is now available at usegalaxy.fr.

Results and impact

By gathering a large variety of tools with standardised input/outputs, W4M’s portal increases LC/GC-MS(MS) and NMR workflows’ efficiency. We highlight how current advances, along with community training as through the yearly international school Workflow4Experimenters, contribute to comprehensive open data analysis practices worldwide.

Label-free quantification of protein post-translational modifications (PTMs) by liquid chromatography coupled to high-resolution mass spectrometry is a powerful approach to reveal PTM-mediated regulatory networks. Nowadays, many robust software are available to analyze the large datasets generated by this technique. To find modified peptides differentially abundant between biological conditions, standard approaches such as Student's t-test can be directly applied on the intensities of the modified peptides. However, applying such approaches does not answer a crucial question: is the difference in intensities of the modified peptides related to the dynamics of their modification, or is it related to the dynamics of the abundance of their associated protein between the compared conditions? This poses statistical problems since the protein dynamics are not directly observed in bottom-up proteomics and missing values complicate the problem. To compare the dynamics of PTMs to their protein, we developed a statistical framework incorporated in a new R package linked to a Shiny app called DiffPTM. It offers functions to compare all quantified PTMs to its reference proteome between multiple conditions. Above all, the DiffPTM app offers a directed data analysis pipeline to produce Excel and PowerPoint files containing multiple graphs for standardized reports.

Thermal proteome profiling (TPP) is a proteome wide technology combining the cellular thermal shift assay with quantitative mass spectrometry to provide insights into protein interactions and states. Statistical analysis of temperature range TPP (TPP-TR) datasets relies on comparing protein melting curves, describing the amount of non-denatured proteins as a function of temperature, between different conditions (e.g. presence or absence of a drug). However, state-of-the-art models are restricted to sigmoidal melting behaviours, while unconventional melting curves represent up to 20% of TPP-TR datasets. We thus propose a novel statistical framework based on hierarchical Gaussian Process models, to make TPP-TR datasets analysis unbiased. The model scaled to multiple conditions and complex TPP-TR protocols. Especially, the analysis of peptide-level TPP-TR datasets, considering melting curves of tryptic peptides instead of protein averages, is implemented using deeper hierarchies. Unbiased analysis of these datasets, of high value for the study of protein post-translational modifications, were yet impossible due to unconventional melting curves abundance. Collectively, this statistical framework extends the analysis of TPP-TR datasets for both protein and peptide level melting curves, offering access to thousands of previously excluded melting curves and paving the way to new biological discoveries on protein interactions, localization and functions.

Unraveling the dynamic interaction between plants and endophytes provides enormous opportunities to improve microbiome-optimized plant growth and health, which makes crops less dependent on fertilizers and pesticides. A critical step here is to unravel the plant biosynthetic pathways involved in the recruitment of endophytes and the regulatory networks governing them. Our project investigates the dynamics and architecture of plant gene regulatory networks (GRNs) to decipher plant biosynthetic pathways by integrative omics strategies. Here, whole transcriptome and metabolomes from roots and root exudates of Arabidopsis thaliana are being investigated to connect expression patterns of genes to metabolites that play a crucial role in the plant-endophytes interactions. To achieve this, we have developed a software, MEANtools, that predicts metabolic pathways, through the integration of transcriptomics with the untargeted metabolomics data and by incorporating knowledge from the known reactions and chemical structures available in the publicly available databases. We are testing the development of MEANtools with high-resolution time-based paired-transcriptomics and -metabolomics datasets, from plants and other species. We further propose that the pathway prediction accuracy of MEANtools can be further enhanced by integrating strategies that allow associating spatial and temporal gene expression with metabolite abundances across samples to identify potentially causal links.

A highly flexible metabolism is manifested in C. albicans wide genetic diversity and phenotypic variation. However, the consequence of this differing genetic background on the metabolic phenotype is not well understood. Set in this context, we are investigating through an integrated metabolomics and lipidomics study, the metabolic consequence of these differences on 24 clinical isolates from five of the most represented clades. Each clade is represented by four isolates with two of these causing high damage while the other two are causing low damage on TR146 (oral epithelial) cell lines. We have optimized the culture and analytical workflow for the metabolome and lipidome profiling of these isolates, in both yeast and hyphae inducing conditions. The workflow includes successive culture in YPD, with the final culture done in both YPD (yeast morphology) and YP (glucose removed to induce transition to hyphae) and at 37 0C, 5% CO2 and 70 rpm. Data processing and features annotation for both polar and lipid compounds are done using an in-house bioinformatics workflow and database. The main results of the analyses will be presented and discussed.

Background: Extracellular vesicles (EVs) mediate inter-organ communication in type 2 diabetes (T2D) pathogenesis. Gut derived EVs (GDE) protein content reflects metabolic state and administering prediabetic GDE induces a diabetogenic phenotype on healthy mice. Analysis of GDE proteomic profile showed an upregulation of acyl-CoA thioesterases and downregulation of rate-limiting enzymes for glycolysis. To unveil the relevance of oral antidiabetic drugs, we hypothesize that metformin and pioglitazone's metabolic actions are dependent on GDE’s proteomic cargo.

Materials and methods: Two groups of mice were fed with either normal chow diet (NCD) or high-fat diet (HFD), then treated with metformin or pioglitazone. GDE were isolated and characterized by nanoparticle tracking analysis. Proteins were extracted and analyzed by nano-LC-MSMS. Statistical analysis was performed using the limma R package.

Results: After treatment, both drugs improved glucose intolerance and liver steatosis compared to prediabetic animals. We identified 159 proteins differentially expressed between HFD and HFD+metformin and 180 between HFD and HFD+pioglitazone and, together with the principal component analysis among groups, these results indicate both drugs alter the GDEs protein composition to resemble NCD.

Conclusion: Metformin and pioglitazone modify GDE-mediated interorgan crosstalk, which plays a role in the progression of dysmetabolism. Modulating this mechanism may have therapeutic implications.

Tandem mass spectra capture fragmentation patterns that provide key structural information about a molecule. Although mass spectrometry is applied in many areas, the vast majority of small molecules lack experimental reference spectra. For over seventy years, spectrum prediction has remained a key challenge in the field. Existing deep learning methods do not leverage global structure in the molecule, potentially resulting in difficulties when generalizing to new data. In this work we propose a new model, MassFormer, for accurately predicting tandem mass spectra. MassFormer uses a graph transformer architecture to model long-distance relationships between atoms in the molecule. The transformer module is initialized with parameters obtained through a chemical pre-training task, then fine-tuned on spectral data. MassFormer outperforms competing approaches for spectrum prediction on multiple datasets, and is able to recover prior knowledge about the effect of collision energy on the spectrum. By employing gradient-based attribution methods, we demonstrate that the model can identify relationships between fragment peaks. To further highlight MassFormer's utility, we show that it can match or exceed existing prediction-based methods on two spectrum identification tasks. We provide open-source implementations of our model and baseline approaches, with the goal of encouraging future research in this area.

Secondary metabolites are a diverse class of small molecules with many applications in medicine and industry. Metabolomics often relies on tandem mass spectrometry to discover novel compounds but interpreting metabolomics mass spectra remains challenging and requires appropriate computational methods.

We developed VarMet, a database search tool for identifying variants of secondary metabolites in mass spectrometry data. This approach extends our previous molDiscovery model for small molecule fragmentation with a modification-tolerant search inspired by VarQuest and adapted to a wide range of metabolites, including polyketides, terpenes, and lipids. We evaluated VarMet on 8,765 spectra from the GNPS spectral library and a chemical database consisting of 44,961 molecules from the NP Atlas and GNPS annotations. Correct variant was ranked first for 35% of heavy (precursor mass >500 Da) and 10% of light spectra (precursor mass <500 Da). For 59% and 25% of the spectra groups the correct variant was within the top ten identifications.

We applied VarMet to the Smenospongia aurea metabolome and demonstrated how it identified a smenamide variant missed by the molecule network approach. Overall, VarMet fills the gap in the identification of novel variants of small molecules and better addresses the chemical diversity of secondary metabolites.

In this paper, we study the problem for finding complex proteoforms from protein databases based on top-down tandem mass spectrum data. An important issue is to handle mass errors of peaks in the input spectrum. In this paper, we propose a new model to handle the mass errors of peaks based on the formulation of the PMG and SMG. Note that the masses of sub-paths on the PMG are theoretical and suppose to be accurate. Our method allows each peak in the input spectrum to have a predefined error range. In the alignment of PMG and SMG, we need to give a correction of the mass for each matched peak within the predefined error range. After the correction, we impose that the mass between any two (not necessarily consecutive) matched nodes in the PMG is identical to that of the corresponding two matched peaks in the SMG. Intuitively, this kind of alignment is more accurate.
We design an algorithm to find a maximum number of matched node and peak pairs in the two (PMG and SMG) mass graphs under the new constraint. Recently, We are working on an improved algorithm that are 3 times faster.

Predicting peptide-MHC binding is important for development of vaccines and immunotherapies. Previously, work on uncovering the dynamics of peptide major histocompatibility complex class I (pMHC-I) interactions have been focused on peptides of 8-11 residues, or 9-mer subsets of longer peptides. However, our examination of large, publicly available mass spectrometry (MS) datasets with a high prevalence of longer peptides (10-14 residues), indicates frequent occurrence of bulging and protrusion events.
Our work uncovers an underappreciated facet of pMHC-I interactions, focusing on longer peptides, using comprehensive MS datasets. We have successfully derived kmer motifs overrepresented in these events and formulated a set of binding rules for longer peptides, which we have validated by experimentally testing designed synthetic peptides for stable pMHC-I binding. Our findings challenge the current paradigm for assessment of binding events and could have implications in the areas of immunology, vaccine development, immunotherapy, and immunoinformatics. Our results provide a novel and quantitative framework for understanding the complex dynamics of long peptide-MHC interactions.

Protein function is highly regulated by the co-ordinated expression and interaction in different tissues and the correct localization within different sub-cellular compartments. A common mechanism for protein regulation is post-translational modification (PTM). Protein phosphorylation is the most prevalent and well-studied PTM in eukaryotes, with involvement in diseases such as cancer and Alzheimer’s. This reversible modification is regulated by two key enzymes; kinases and phosphatases. Despite their importance, the available kinase-substrate data on phosphorylation and its biological context (e.g. tissue localization and pathological association) is very limited. However, with kinase-substrate association, important pharmacological targets, better context-aware information on protein phosphorylation and kinase-substrate interaction and functional association is essential to define their role in cellular communication and disease. In this work, we show our initial step to identify and extend kinase-substrate association based on different biological and pathological context by re-processing and re-using available public mass-spectrometry based proteomics data. Identified kinase-substrate association and protein phosphorylation information are integrated with functional and physical protein association networks, tissue and sub-cellular annotations, and various sequence, structural and biophysical features. In summary, our work on identifying tissue specific phosphoproteins and phosphosites show our initial steps towards understanding differential phosphorylation across different sub-cellular location and tissue types.

Nonribosomal peptides (NRPs) represent a medically-important class of secondary metabolites. Evolutionary close microorganisms often produce structurally similar NRPs that form so-called NRP families. While many NRP families are already described, most of them include only a few representative compounds. NRPs are synthesised by special enzymes encoded in biosynthetic gene clusters (BGCs) that challenge NRP discovery directly from genomic data.
Here we present Nerpa-MS, a tool that integrates genomics and mass spectrometry data for high-throughput NRP discovery. Nerpa-MS matches a given BGC against the database of known NRPs, uses the best hits as templates to generate many putative NRP structures, and filters out erroneous predictions using tandem mass spectra. Template-based generation allows Nerpa-MS to discover NRPs with complex chemical structures and containing rare building blocks, which are missed by the existing metabologenomics approaches. We benchmarked our tool on public paired genomics-metabolomics datasets and demonstrated its ability to enrich the currently scarce NRP families.