Melanie Vollmar

The Protein Data Bank in Europe (PDBe; pdbe.org) provides open access to experimentally determined biomacromolecular structures and enriches these data with biological and functional context. Here, we describe recent developments that expand the information connected to PDB structures and introduce new ways for researchers and computational tools to access it. Literature integration has been extended to additional open-access publications, substantially increasing the availability of residue-level annotations mined from the scientific literature. Cross-links to molecular dynamics simulations in MDposit have been established, and simulation-derived residue-level annotations, including relative solvent accessibility and membrane-related properties, have been integrated into PDBe-KB (pdbekb.org). PDBe entry pages now expose additional experimental and sequence information, links to related resources, and extended PDB identifiers. Protein channel, tunnel, and pore analyses generated using MOLE have also been incorporated into our weekly release pipeline and made available for download. Improved API documentation facilitates the discovery of data fields and search parameters, while three Model Context Protocol servers facilitate agentic access to the PDBe APIs, Search, and the PDBe-KB knowledge graph. Together, these developments make structural data and its wider biological context easier to discover, interpret, and reuse through both conventional and agentic workflows.

Mechanistic understanding of viral protein-protein interactions enables global health security and pandemic preparedness, yet experimental characterisation remains difficult, costly, and often restricted to specialised laboratories. Here, we describe an in silico campaign using AlphaFold2 and AlphaFold-Multimer to predict monomers and dimer structures within 2,812 viral proteomes from 23 viral families relevant to human health. We report high-confidence predicted structures of 5,279 hetero-, and 2,749 homo-dimers. Structural clustering of interfaces reduces the set fivefold to 1,598 consolidated by form and function, of which 471 (29.5%) lacked detectable similarity to experimentally determined interfaces in the PDB. These clusters yielded two notable findings: new data on vaccinology-relevant glycoproteins, and accurate prediction of the specificity and cleavage-site recognition of viral proteases, which are targets for anti-viral development. Viral monomers and high-confidence dimers are available for interactive browsing in the Pandemic Preparedness Portal of the AlphaFold Database ( https://alphafold.ebi.ac.uk/ ), and the full dataset is available for bulk download.

Protein structures are crucial in understanding the function, mechanism and disease-causing variants of proteins within any living cell. A number of experimental techniques are employed by researchers to determine such structures. Through structure inspection in molecular viewers, combined with supporting biochemical and biophysical experiments, scientists are able to identify the function of a protein, its reaction mechanism and effects caused by sequence variation. These detailed findings, supported by experimental results, are documented by being described in the scientific literature and by making the accompanying data open source. However, it has become increasingly difficult for a reader, in particular a non-expert, to access the correct additional information and assess the validity of the conclusions drawn based on experimental results. A reader is often required to resort to a number of different software packages to access the different data types. Here, we present a first-of-its-kind implementation of an artificial intelligence- and text-mining-supported software tool that allows the association of mentions in the text of one or more specific protein residues with their corresponding counterparts in the respective protein structure or structures. Our application allows a researcher to explore a residue of interest in the context of a publication and its respective protein structure, supported by its experimental evidence, in a single view. We describe model implementation, annotation extraction, downstream processing, dissemination and visualization at the IUCr and PDBe. The application presented is primarily aimed at readers of IUCr publications and users visiting the PDBe entry pages. However, we believe that in the future our application will be a valuable tool for reviewers of new submissions to IUCr journals and may even be useful as a curation tool involving the authors of a publication as annotation validators.

Protein Data Bank in Europe (PDBe) is a founding member of the worldwide Protein Data Bank (wwPDB), delivering open access to experimentally determined macromolecular structures. PDBe also delivers enriched annotations contributed by the PDBe-Knowledge Base (PDBe-KB) consortium. The macromolecular entry pages are the primary interface for millions of users who explore experimental structure data. Here, we describe the redesign of the PDBe entry pages that organize content into logical views, thereby improving usability and facilitating the use of structure models, driving fundamental and applied research, and supporting education. The new design introduces several enhancements, including integrated and central 3D visualization, sequence feature exploration, standardized molecular scenes, AI-driven residue-level annotations from text mining of the literature, and streamlined annotation access via refactored APIs. Importantly, researchers can now upload their own residue-level features and visualize them directly in structural context, displayed alongside PDBe-KB annotations. This functionality enables scientists to interpret unpublished or private data in relation to high-quality structural information, lowering barriers for those without prior expertise in structural biology. Together, these updates create a more accessible, flexible, and scalable framework for interacting with structural data, expanding the resource’s value to both domain specialists and the wider life sciences community.

Human methionine synthase (MTR) is an essential enzyme of one carbon metabolism. Consisting of a catalytic N-half and a cobalamin binding C-half, MTR utilises this intricate organometallic cofactor in the methyl transfer from methyltetrahydrofolate to homocysteine producing methionine. Cobalamin loading into MTR, and its subsequent activation, requires methylmalonic aciduria and homocystinuria Type D (MMADHC) protein and methionine synthase reductase (MTRR), respectively. However, the molecular basis of cobalamin binding and activation of human MTR aided by MMADHC and MTRR remains unknown. Here, using cryo-electron microscopy, we determined structures of human MTR in its apo, and cobalamin bound states. Apo MTR adopts a conformation where the two halves of the enzyme act independently with the C-half posed to bind cobalamin. Binding of cobalamin and its activation causes conformational changes in MTR that result in a flexible catalytically active state. AlphaFold predictions, validated by interaction studies, show that MMADHC interacts with the C-half of apo MTR to facilitate cobalamin loading. Unexpectedly we found that MTRR interacts at two distinct sites within the C-half of MTR which may aid in activation. Collectively these findings lay the groundwork to uncover the mechanisms through how MMADHC and MTRR coordinate cobalamin loading and activation of human MTR.

Protein structures are crucial in understanding function, mechanism and disease-causing variants of proteins within any living cell. A number of experimental techniques are employed by researchers to determine said structure. Through structure inspection in molecular viewers combined with supporting biochemical and biophysical experiments, scientists are able to identify a protein’s function, reaction mechanism and effects caused by sequence variation. These detailed findings supported by experimental results are documented and described in detail in scientific literature and by open sourcing the accompanying data. By writing a detailed report about the findings and providing evidence in additional files and complementary data formats it has become increasingly difficult for a reader, in particular a non-expert, to access the correct additional information and assess the validity of the drawn conclusion based on experimental results. It often requires a reader to resort to a number of different software packages to access the different data types. Here, we present a first-of-its-kind implementation of an artificial intelligence and text mining supported software tool that allows linking of text mentions of specific protein residues to their corresponding counterpart in the respective protein structure. An identified residue is highlighted in the publication text and upon interaction with the annotation, a molecule viewer displays the associated protein structures in the publication which contain said residue. The viewer is complemented by a display table that contains protein structure quality metrics for each occurrence of a residue. As such a reader can now explore a residue of interest they are currently assessing in a publication within its respective protein structure supported by its experimental evidence in a single view and application.

Join us for a live panel discussion as part of our ongoing webinar series, Large Language Models and their applications in bioinformatics. This interactive QA session offers attendees the opportunity to engage with speakers from the series, ask new questions, and revisit key discussions.

As large language models (LLMs) continue to revolutionise artificial intelligence applications, the importance of high-quality data preparation has never been more critical. This webinar dives into the art and science of preparing datasets for effective LLM training, offering actionable insights for AI practitioners, data scientists, and engineers. We will explore the end-to-end process of data preparation, beginning with data collection strategies and progressing through cleaning, preprocessing, tokenisation, and annotation. Emphasis will be placed on identifying and mitigating biases, managing multilingual datasets, and ensuring data quality and diversity to enhance model performance. Real-world case studies will illustrate common pitfalls and solutions, while hands-on demonstrations will provide practical techniques for optimising datasets. Participants will gain a deeper understanding of how well-structured and curated data can significantly impact an LLM’s capabilities, reduce training costs, and improve ethical AI outcomes. Whether you are building LLMs from scratch or fine-tuning existing models, this session will equip you with the knowledge to leverage your data assets effectively. Join us to unlock the potential of data preparation and enable your LLMs to achieve unparalleled performance and generalisation.

Sharing your code is a good first step towards building open source software, but it is not enough to ensure its success. This webinar focuses on practical issues to increase impact of open software projects, including maintenance, community building, discoverability, project management, documentation, licensing, and user needs.

We present a novel system that leverages curators in the loop to develop a dataset and model for detecting structure features and functional annotations at residue-level from standard publication text. Our approach involves the integration of data from multiple resources, including PDBe, EuropePMC, PubMedCentral, and PubMed, combined with annotation guidelines from UniProt, and LitSuggest and HuggingFace models as tools in the annotation process. A team of seven annotators manually curated ten articles for named entities, which we utilized to train a starting PubmedBert model from HuggingFace. Using a human-in-the-loop annotation system, we iteratively developed the best model with commendable performance metrics of 0.90 for precision, 0.92 for recall, and 0.91 for F1-measure. Our proposed system showcases a successful synergy of machine learning techniques and human expertise in curating a dataset for residue-level functional annotations and protein structure features. The results demonstrate the potential for broader applications in protein research, bridging the gap between advanced machine learning models and the indispensable insights of domain experts.