EuBIC-MS Winter School 2024 – Full Program

Challenges in Computational (clinical) Lipidomics and QC – a tale of impostors, cut-throat competition and lack of etiquette

Lipids play an integral part in biological functions and emerge as crucial biomarkers in medical research, notably in predicting cardiovascular risks. However, their analysis via mass spectrometry poses challenges due to their structural diversity, concentration ranges, isomerism, and complex fragmentation. This presentation highlights pitfalls in lipid identification, emphasizing the need for evidence-backed reporting using standardized nomenclature and lipid-class-specific MS features, together with class-specific quantification. Moreover, the talk explores strategies and presents tools for enhancing reporting practices in lipidomics experiments to foster reanalysis and comparability. Presenting ongoing work by the HUPO-PSI QC group, I introduce the mzQC file format and QC metrics tailored for mass spectrometry, offering insights into their application in lipid studies. Drawing from experiences in an international clinical lipidomics ring trial, I delve into the hurdles faced and strategies employed to establish concentration ranges measured by multiple labs across the world for four clinically relevant ceramides, addressing key challenges encountered during setup, execution, and analysis.

What determines precision and accuracy of DIA proteomics?

DIA proteomics has long been viewed as a discovery method with
good label-free quantitative performance. However, while algorithms for peptide
identification from DIA data have progressed immensely in the past years,
quantification approaches have in contrast remained rather primitive. In this talk, I will
discuss recent technology and algorithm developments towards boosting precision
and accuracy of DIA.

Making proteomics data FAIR: Challenges and rewards

First of all I will summarise the current state of the art with regards to open data practices in proteomics, which have revolutionised the field in recent years. Throughout the talk, I will cover some key questions: why is good to make data available in the public domain? how this can be done?, and maybe even more importantly, what are these practices useful for? I will then highlight some nice examples of how this data is being reused by the community. Finally, I will explain some of the upcoming challenges.

Special Session: Scientific communications in the age of skibidi toilet

This 3 part workshop by Ben Neely and Ben Orsburn will begin with a history of the terrible and inaccurate named “News In Proteomics Research” blog as described by the blog author. Some chronological history will help to portray some of the perils of social media and how these can have long lasting negative effects on your career. The second part of this informal workshop will be focused on how to set up and run your own podcast series. Part 3 will feature the distortion of social media with audience participation to let us all learn something. (This description of this workshop was written by Ben Orsburn)

A principled approach to process, analyse and interpret single-cell proteomics data

Mass spectrometry (MS)-based single-cell proteomics (SCP) has become a credible player in the single-cell biology arena. Continuous technical improvements have pushed the boundaries of sensitivity and throughput. However, the computational efforts to support the analysis of these complex data have been missing. Strong batch effects coupled to high proportions of missing values complicate the analysis, causing strong entanglement between biological and technical variability. We propose a simple, yet powerful approach to address this need: linear models. We use linear regression to model and remove undesired technical factors while retaining the biological variability, even in the presence of high proportions of missing values. The key advantage of linear models lies in the interpretability of the results they generate. Inspired by previous research, we streamlined modelling and exploration of the patterns induced by known technical and biological factors. The exploration enables a thorough assessment of the model coefficients, and highlights key factors influencing SCP experiments. Further exploration of the unmodelled variance recovers unknown but biologically relevant patterns in the data, leveraging the power of single-cell proteomics technologies. We successfully applied our approach to a diverse collection of SCP datasets, and could demonstrate that it is also amenable for integrating datasets acquired using different technologies. Our approach represents a turning point for principled SCP data analysis, moving the tension point from how to perform the analysis to result generation and interpretation.

AlphaPeptDeep: a deep learning framework for peptide property prediction and what we can do with it

Deep learning has demonstrated its efficacy in enhancing the search capabilities of mass spectrometry (MS)-based proteomics data. Leveraging our AlphaX ecosystem, we have developed a deep learning framework called AlphaPeptDeep, equipped with an intuitive programming interface for both training and transfer learning. This framework is designed to handle diverse properties from peptide sequences with sufficient training data. Our studies revealed that AlphaPeptDeep excelled in predicting fragments (MS2), retention times (RT), and ion mobilities (IM) of peptides. Moreover, transfer learning significantly enhanced the accuracy of these predictions, even when trained on a limited dataset specific to certain experimental conditions. In addition to its proficiency in MS2/RT/IM prediction, AlphaPeptDeep can also forecast Major Histocompatibility Complex (MHC)-binding peptides. This feature streamlines the direct search of data-independent acquisition (DIA) MS data without the necessity of data-dependent acquisition (DDA)-based spectral libraries for immunopeptides. During this presentation, we will delve into other peptide properties, such as MS-detectability and charge states, which AlphaPeptDeep can predict through deep learning. We will also highlight the simplicity of model training using AlphaPeptDeep.

W.T.F. are isotopes

In this overview presentation, we will provide an introduction to elemental isotopes. What are they, how were they discovered, and how are these isotopes incorporated in molecular analytes? The discovery of stable isotopes changed chemistry forever as we had to rethink the concept of molecular mass. Therefore, we determine molecules by their monoisotopic and average masses. The mass spectrometry community found an interest in how these isotopes were distributed to aid in molecular identification. This identification worked well for low-mass molecules as it is easy to calculate a theoretical isotope distribution for a given molecule. However, for larger molecules, the calculation of the isotope distribution becomes cumbersome and requires a deep understanding of mathematics and computer science. Therefore, we explain the history of algorithmic design to calculate the isotope distributions. Furthermore, we introduce concepts like, e.g., isotopologues, the aggregated isotope distribution, and the fine isotope distribution.

Rethinking the space race in proteomics informatics; are we using the right metrics?

Machine learning for predicting the LC-MS behaviour of peptides have become widespread in the field of computational proteomics.  These predictions find utility across various applications such as experimental design, rescoring, spectral library creation, and many more. Despite the improvement in commonly used metrics (such as number of PSMs or quantified proteins) when incorporating machine learning, there is disagreement on metrics to quantify this improvement. For example, when new models get introduced their prediction accuracy in terms of the spectral angle or Pearson correlation is compared with the current state-of-the-art. However, making a fair comparison is difficult and does not always help your chances of getting published (unfortunately). Furthermore, while an improvement in prediction accuracy or more identified PSMs can indicate a potential to better answer biological questions, in many cases more accurate predictions will not influence the quality of answers to our biological questions. Are we hyper focused on the accuracy of our models and are we thus wasting valuable research time? In this talk I will present three practical cases where metrics the field commonly uses are not indicative of the quality of answers to biological questions.

Simplified and Automated Analysis of Large-Scale Proteomics Datasets

Analysis of large-scale proteomics dataset involves several steps including quality control (QC), search, quantitation, visualization and reporting. Here, we demonstrate implementing these systems using an easy-to-use, interactive, customizable, web-based next-generation analysis platform for deeper proteomics insights. The highly scalable platform supports standard and custom workflows and robust pipelines, enables reproducible and value driven data transformation and, data visulalzation by levraging  supercomputing for higher throughput. The talk will cover the following aspects: 1) User interface with the option to be standardized or customized for diverse project needs 2) Ability to plug in any tool to play with complex web of dynamic pipelines and 3) visulaization and reporting of proteomics datasets in to identify biological patterns and trends.

Thomas Burger

CNRS, FR

Thomas Burger is a CNRS senior scientist specialized in statistical and computational methodologies to improve knowledge extraction from high-throughput mass spectrometry based proteomics data. He holds two MS degrees in computer sciences and in applied mathematics (in 2004), a PhD in pattern recognition (in 2007) and a Habilitation thesis (in 2017), all from Grenoble Alpes University (France). Thomas has been an associate professor in machine learning with South Brittany University for three years, before rushing back to his beloved mountains in 2011, as a CNRS scientist. Since then, he has been affiliated with EDyP, a joint lab with Grenoble Alpes University, CNRS, CEA and INSERM. His research group is essentially focused on theoretical questions underlying missing value imputation, multi-omics data fusion and false discovery rate control, while maintaining and developing the Prostar software suite for the statistical analysis of label-free proteomics data

Cecilia Lindskog

Uppsala University, SE

Cecilia Lindskog is Associate Professor in experimental pathology and head of her research group at the Faculty of Medicine at Uppsala University in Sweden. Her research focuses on integrating transcriptomics and antibody based proteomics with the purpose of linking cell type specificity with the function and disease mechanisms. Additionally, she is leading the tissue-based profiling of Human Protein Atlas.

Ben Neely

NIST, US

Ben Neely is a research chemist with the National Institute of Standards and Technology in Charleston, South Carolina (NIST-Charleston) focused on analytical biochemistry and bioinformatics. His diverse background includes microbiology, wildlife disease, cancer biology, biomarker discovery and validation (protein and glycan) and bioinformatics. His focus is primarily bottom-up proteomic methods and data analysis (DDA and DIA, and metaproteomics). He leads the Comparative Mammalian Proteome Aggregator Resource (CoMPARe) Program, generating standardized proteomic data across non-model species. Approaching this problem requires genome sequencing and annotation, understanding optimum search space construction, and implementing quality control metrics with and without known fasta (including generative ML applications). These solutions have helped drive techniques to identify unknown species, and to apply bottom-up proteomics in complex samples such as host-virus-vector systems.

Nils Hoffmann

FZ Jülich, DE

Nils studied computer science and developed an interest in biology, chemistry, and mass spectrometry during his time at Bielefeld University. After completing his PhD on processing GCxGC-MS metabolomics data, he transitioned to the IT industry, where he contributed to maintaining and developing search engines for the European Patent Office. Nils later returned to the scientific field, joining ISAS Dortmund and currently coordinates activities for the de.NBI LIFS consortium and works at Forschungszentrum Jülich, sharing duties in the development and operations of the de.NBI Cloud portal and coordinating the German ELIXIR node activities related to cloud computing and interoperability.

Vadim Demichev

Institute of Biochemistry, Charité, DE

Vadim Demichev’s laboratory for Quantitative Proteomics, Charité –
Universitätsmedizin Berlin, works on development and application of new mass
spectrometry methods, to boost speed, sensitivity, reliability and quantitiative
accuracy of proteomics. Our recent developments include the QuantUMS concept for
machine-learning guided quantification in proteomics, the plexDIA platform for
multiplexed data-independent acquisition proteomics and the Slice-PASEF
technology that maximises the MS/MS duty cycle and hence the sensitivity and
precision of the mass spectrometry method. We also develop the DIA-NN algorithm
suite for proteomics data processing.

Juan Antonio Vizcaíno

EMBL-EBI, UK

Dr. Juan Antonio Vizcaíno is leading the Proteomics Team at the European Bioinformatics Institute (EMBL-EBI, Cambridge, UK). His group is responsible of the development of the PRIDE database (https://www.ebi.ac.uk/pride/), the world-leading public repository for mass spectrometry (MS) proteomics data, and related tools and resources. In addition, he co-founded and is coordinating the ProteomeXchange Consortium (http://www.proteomexchange.org/), standardizing data submission and dissemination in proteomics resources worldwide. He actively promotes open data policies in the proteomics field and has participated in many studies where public proteomics datasets are reused for different purposes. Additionally, over the years, he has heavily contributed to the development of open proteomics data standard formats and related software, under the umbrella of the HUPO Proteomics Standards Initiative (PSI). He is also co-leading the ELIXIR Proteomics Community (https://elixir-europe.org/communities/proteomics) in Europe. Furthermore, he has co-organised the annual Proteomics Bioinformatics course at EMBL-EBI since 2009.

Ben Orsburn (Special Guest)

Johns Hopkins University School of Medicine, US

Ben Orsburn is a Principal Investigator at the Johns Hopkins University School of Medicine in the Department of Pharmacology. His research centers on measuring atypical biomolecules of interest to human health. He spends most of his free time engaged in scientific communications through projects like the News In Proteomics blog, the Proteomics Show podcast, and as chair of the US HUPO Virtual Media Organization.

Laurent Gatto

De Duve Institute / UCLouvain, BE

Laurent is an Associate Professor of Bioinformatics at the de Duve Institute, UCLouvain, in Belgium. His research group focuses on the development and application of statistical learning for the analysis, integration and comprehension of large scale biological data. The development and publication of scientific software is an integral part of the labs work, as reflected by their numerous contributions to the Bioconductor project. Laurent is an avid open and reproducible research advocate, making his research outputs openly available. He is a Software Sustainability Institute fellow, a Data and Software Carpentry instructor and a member of the Bioconductor technical advisory board.

Wen-Feng Zeng

MPI Biochemistry, DE

Dr. Wen-Feng Zeng got his PhD in pFind Lab in Institute of Computing Technology, Chinese Academy of Sciences. After he worked as an assistant researcher in pFind Lab, he joined Mann Lab in Max-Planck Institute of Biochemistry as a postdoc researcher. His research interests include computational proteomics/glycoproteomics and deep learning in proteomics.

Dirk Valkenborg

Hasselt University, BE

Dirk Valkenborg is team leader of BIER-lab (Bioinformatics, Intelligence, Exploration and Research) at Hasselt University where he is affiliated to the Data Science Institute and the Center for Statistics. With his education in engineering, biostatistics, and mathematics and an interest in biology and clinical research, he develops theories and applications for today’s problems in biotechnology and human health. His research is centered on data processing, statistical analysis of various ‘omics’ data, and integrating data workflows.

Robbin Bouwmeester

Ghent University / VIB, BE

Robbin is a researcher dedicated to studying peptide and small molecule behaviour in liquid chromatography, ion mobility, and mass spectrometry. He completed his PhD in 2020 as part of the H2020 project MASSTRPLAN. Following his PhD, Robbin took on the role of a postdoctoral scientist at Johnson & Johnson and the Flemish Institute for Biotechnology (VIB). There, he applied his knowledge to the development of machine learning models designed to improve quality control processes in pharmaceutical workflows.

Niveda Sundararaman

Cedars-Sinai, US

Niveda Sundararaman is the Bioinformatics lead at the Van Eyk Lab at Cedars Sinai Medical Center, Los Angeles, CA. She carried out her M.Sc. in bioinformatics and computation at Georgia Tech. Her focus is primarily in DIA bottom-up proteomic data analysis through the development of bioinformatics algorithms, pipelines and visualization tools to handle large-scale mass spectrometry datasets.