Biomedical knowledge about the brain increases every day, with a rapidly growing number of scientific publications, datasets, and software tools. While this informational plethora is not merely comprehensible by human beings, recent developments in information science and computational linguistics aim to make this knowledge programmatically accessible by literature mining. However, integrating these semantic methods into neuroimaging standards remains insufficient, hindering researchers from unraveling their full potential. Therefore, we developed the semantic meta-analysis platform The Virtual Brain adapter of semantics (TVBase) that projects biomedical knowledge preserved in over 36 million scientific articles onto a 3D standardized brain. The literature-mining platform SCAIView was used to extract ontologically defined biomedical entities and their associations with brain anatomy from the PubMed database. By querying a specific concept, the association strength with each anatomical term was calculated using entropy. To project the data onto a standardized brain, we created a unique transformation matrix that links over 800 anatomical terms to voxel coordinates of a parcellated standard brain. This novel method of knowledge projection extracts region-specific information about biomedical concepts from the literature to support translational multi-scale approaches to computational neuroscience. The multi-purpose software framework TVBase is openly available as a Python library. It aims for hypothesis-free neuroimaging pattern interpretation, hypothesis generation, and applications in personalized medicine.

IntroductionTranslational biomedical research is increasingly collaborative and multimodal, making secure, high-quality data capture, curation, and analytics a major challenge. This work aims to provide an overview of existing medical research data platforms to support informed platform selection for translational biomedical research.MethodsAs part of an ongoing Fraunhofer Request for Proposal (RFP) process, we developed a requirements assessment tool for users across the Fraunhofer ecosystem. In parallel, we compiled a structured overview of medical research data platforms through an open collaboration between academic and industry experts, who supplemented our market screening by identifying additional relevant platforms. Using a standardized questionnaire on key aspects of distributed data collaboration, we collected harmonized platform descriptions and organized them into a side-by-side overview with an accompanying feature weighting matrix.ResultsThe study yielded a structured, comparative characterization of medical research data platforms across five functional classes, highlighting common strengths in security, interoperability, data quality, and multimodal data support. We devised (developed) a platform feature-partner weight matrix that enables context-sensitive platform scoring without imposing a predefined global ranking. In this way, users can align platform scoring with their specific translational research requirements.DiscussionThis structured, overview is intended to accelerate decision-making in the medical research community when choosing data platforms. By supporting context-sensitive, feature-weighted selection rather than one-size-fits-all comparisons, it acknowledges diversified research needs and can be updated as technologies and practices evolve.

Neurological complications are among typical clinical manifestations, both during the acute phase of SARS-CoV-2 infection and long-term post-acute COVID-19 sequelae, i.e., Long COVID. A growing body of evidence connects SARS-CoV-2 infection to systemic inflammation, peripheral immune dysregulation, functional alteration of the blood–brain barrier (BBB), neuroinflammatory processes, mitochondrial dysfunction and pathological protein aggregation, all of which are central to the pathogenesis of major neurodegenerative diseases. In this review, we discuss the converging shared underlying molecular mechanisms that link the ‘acute’ SARS-CoV-2 infection that often only lasts for days and ‘chronic’ neurodegenerative diseases that often take decades to develop. We further review human induced pluripotent stem cells (iPSCs)-derived brain organoid and BBB models for Parkinson’s disease (PD) and Alzheimer’s disease (AD) as well as artificial intelligence (AI)-facilitated knowledge graph methods that are used to hypothesize and validate the potential mechanistic links. We propose that besides the established links between viral infections and neurodegenerative or neuroinflammatory conditions, such as Epstein-Barr virus (EBV) in multiple sclerosis or herpes zoster in AD, SARS-CoV-2 infection might be another important factor driving the future burden of the most common neurodegenerative diseases.

Biomedical artificial intelligence (AI) systems increasingly extract, organize, and reuse scientific claims from literature, clinical trials, and regulatory documents. But automatic extraction alone does not make a claim reliable evidence: a claim becomes useful only when it can be traced to its source, linked to the quantitative details that support it, and read within its biomedical context and uncertainty. This matters as large language models (LLMs) and increasingly autonomous systems drive evidence synthesis, knowledge graph (KG) construction, and decision support. Many text-mining and LLM pipelines remain relation-centric: they capture entities and relations such as Drug–TREATS–Disease, but drop the dose, effect size, population, comparator, uncertainty, and conditions under which a claim holds. Such relations can look actionable yet remain hard to verify, compare, or reuse. In this perspective, we argue for a shift toward quantitative evidence mining—extracting values, units, measured entities and properties, context, uncertainty, provenance, and plausibility as structured evidence units that populate evidence-aware KGs and can be checked for source grounding, unit consistency, completeness, and biological plausibility. We outline a framework for plausibility-aware AI that treats extracted claims not as final answers but as auditable evidence objects, making clear what was measured, how much it changed, in which setting, with what uncertainty, and from which source. The central risk is not only incorrect extraction, but claims that look like evidence while lacking the structure needed to trust them.

Clinical guidelines typically focus on single diseases and often overlook the complex comorbidities emerging in the aftermath of COVID-19. As growing evidence links COVID-19 to long-term neurological outcomes, integrative approaches are required to uncover cross-disease relationships. We present a graph-based framework that combines clinical trial data, biomedical mechanisms, and guideline recommendations to support comorbidity-aware care. Our approach comprises two components: (1) a semantic analysis of clinical guidelines for COVID-19, Parkinson’s disease, Alzheimer’s disease, and Multiple Sclerosis (MS) using MPNet-based sentence embeddings and UMAP; and (2) the construction of a comorbidity knowledge graph (KG) in Neo4j that integrates clinical trials, mechanistic hypotheses, and guideline content enriched with ontological annotations. Semantic clustering reveals distinct disease-specific patterns; however, when condition names are masked, neurological disorders begin to overlap while COVID-19 remains isolated, exposing thematic gaps in current guidance. The KG further identifies intermediates such as stroke and olfaction disorders that link COVID-19 to neurodegenerative diseases, highlighting under-recognized comorbidities. This study lays the groundwork for AI-driven, comorbidity-aware clinical decision support. By aligning clinical and mechanistic evidence, it enables adaptive and explainable tools, such as Retrieval-Augmented Generation, to assist clinicians and policy-makers in navigating complex multi-condition scenarios.

Background The COVID-19 pandemic has intensified concerns about its long-term neurological impact, with evidence linking SARS-CoV-2 infection to neurodegenerative diseases (NDDs) such as Alzheimer’s (AD) and Parkinson’s (PD). Patients with these conditions face higher risk of severe COVID-19 outcomes and may experience accelerated cognitive or motor decline following infection. Proposed mechanisms—including neuroinflammation, blood–brain barrier (BBB) disruption, and abnormal protein aggregation—closely mirror core features of neurodegenerative pathology. However, current knowledge remains fragmented across text, figures, and pathway diagrams, limiting integration into computational models that could reveal systemic patterns. Results To address this gap, we applied GPT-4 Omni (GPT-4o), a multimodal large language model (LLM), to extract mechanistic insights from biomedical figures. Over 10,000 images were retrieved through targeted searches on COVID-19 and neurodegeneration; after automated and manual filtering, a curated subset was analyzed. GPT-4o extracted biological relationships as semantic triples, grouped into six mechanistic categories—including microglial activation and barrier disruption—using ontology-guided similarity and assembled into a Neo4j knowledge graph (KG). Accuracy was evaluated against a gold-standard dataset of expert-annotated images using Biomedical Bidirectional Encoder Representations from Transformers (BioBERT)–based semantic matching. This evaluation enabled prompt tuning, threshold optimization, and hyperparameter assessment. Results show that GPT-4o successfully recovers both established and novel mechanisms, yielding interpretable outputs that illuminate complex biological links between SARS-CoV-2 and neurodegeneration. Conclusions This study demonstrates the potential of multimodal LLMs to mine biomedical visual data at scale. By complementing text mining and integrating figure-derived knowledge, our framework advances understanding of COVID-19–related neurodegeneration and supports future translational research.

Data Harmonization is an important yet time-consuming process. With the recent popularity of applications using Language Models (LMs) due to their high capabilities in text understanding, we investigated whether LMs could facilitate data harmonization for clinical use cases. To evaluate this, we created PASSIONATE, a novel Parkinson's disease (PD) variable mapping schema as a ground truth source for pairwise cohort harmonization using LLMs. Additionally, we extended our investigation using an existing Alzheimer's disease (AD) CDM. We computed text embeddings based on two language models to perform automated cohort harmonization for both AD and PD. We additionally compared the results to a baseline method using fuzzy string matching to determine the degree to which the semantic capabilities of language models can be utilized for automated cohort harmonization. We found that mappings based on text embeddings performed significantly better than those generated by fuzzy string matching, reaching an average accuracy of over 80% for almost all tested PD cohorts. When extended to a further neighborhood of possible matches, the accuracy could be improved to up to 96%. Our results suggest that language models can be used for automated harmonization with a high accuracy that can potentially be improved in the future by applying domain-trained models.

The COVID-19 pandemic has generated a vast volume of research, yet much of it focuses on individual diseases, overlooking complex comorbidity relationships. While extensive literature exists on both neurodegenerative diseases (NDDs), such as Alzheimer's and Parkinson's, and COVID-19, their intersection remains underexplored. Co-morbidity modeling is crucial, particularly for hospitalized patients often presenting with multiple conditions. This study investigates the interplay between COVID-19 and NDDs by integrating knowledge graphs (KGs) built from curated biomedical datasets and text mining tools. We performed comprehensive analyses-including path analysis, phenotype coverage, and mapping of cellular and genetic factors-across multiple KGs, such as PrimeKG, DrugBank, OpenTargets, and those generated via natural language processing (NLP) methods. Our findings reveal notable variability in graph density and connectivity, with each KG offering unique insights into molecular and phenotypic links between COVID-19 and NDDs. Key genetic and inflammatory markers, especially immune response genes, consistently appeared across graphs, suggesting a shared pathogenic basis. By unifying structured biological data with unstructured textual evidence, we enhance co-morbidity modeling and improve recall in identifying mechanisms underlying COVID-19-NDD interactions. This integrative framework supports the development of a co-morbidity hypothesis database aimed at facilitating therapeutic target discovery. All data, methods, and instructions for accessing the co-morbidity hypothesis database are publicly available at: http s://github.com/SCAI-BIO/covid-NDD-comorbidity-NLP.

Summary COVID-19 is increasingly recognized as a potential trigger or accelerator of neurodegenerative diseases such as Alzheimer’s and Parkinson’s. To systematically explore the putative molecular and clinical associations between them, we present CBM KG (Causal Biological Mechanisms Knowledge Graph)—a manually curated, comorbidity-centric resource developed within the EU-funded COMMUTE project. CBM KG integrates over 2,800 cause-and-effect or correlative relationships from 63 peer-reviewed publications, highlighting key mechanisms such as viral entry routes, blood-brain barrier alteration, microglial activation, neuroinflammation, and APOE ε4-associated susceptibility. Each relationship in the graph is fully traceable to its source evidence, ensuring transparency and reproducibility. Unlike general-purpose or single disease-focused knowledge graphs, CBM KG is specifically designed to represent causal biological mechanisms spanning both infectious and neurodegenerative processes. By encoding directional, cause-and-effect relationships, it supports the interpretation of clinical co-occurrences through plausible mechanistic links between overlapping disease pathways, offering high-resolution insights at both molecular and clinical levels. Availability and implementation The BEL files, Neo4j database, and Cytoscape visualization files are publicly available at: . ### Competing Interest Statement The authors have declared no competing interest. European Health and Digital Executive Agency (HADEA), 101136957

Motivation:Drug repurposing is gaining interest due to its high cost-effectiveness, low risks, and improved patient outcomes. However, most drug repurposing methods depend on drug-disease-target semantic connections of a single drug rather than insights from drug combination data. In this study, we propose SynDRep, a novel drug repurposing tool based on enriching knowledge graphs (KG) with drug combination effects. It predicts the synergistic drug partner with a commonly prescribed drug for the target disease, leveraging graph embedding and machine learning (ML) techniques. This partner drug is then repurposed as a single agent for this disease by exploring pathways between them in the KG. Results:HolE was the best-performing embedding model (with 84.58% of true predictions for all relations), and random forest emerged as the best ML model with an area under the receiver operating characteristic curve (ROC-AUC) value of 0.796. Some of our selected candidates, such as miconazole and albendazole for Alzheimer's disease, have been validated through literature, while others lack either a clear pathway or literature evidence for their use for the disease of interest. Therefore, complementing SynDRep with more specialized KGs, and additional training data, would enhance its efficacy and offer cost-effective and timely solutions for patients. Availability and implementation:SynDRep is available as an open-source Python package at https://github.com/SynDRep/SynDRep under the Apache 2.0 License.

Disease surveillance systems play a crucial role in monitoring and preventing infectious diseases. However, the current landscape, primarily focused on fragmented health data, poses challenges to contextual understanding and decision-making. This paper addresses this issue by proposing a semantic framework using ontologies to provide a unified data representation for seamless integration. The paper demonstrates the effectiveness of this approach using a case study of a COVID-19 incident at a football game in Italy. In this study, we undertook a comprehensive approach to gather and analyze data for the development of ontologies within the realm of pandemic intelligence. Multiple ontologies were meticulously crafted to cater to different domains related to pandemic intelligence, such as healthcare systems, mass gatherings, travel, and diseases. The ontologies were classified into top-level, domain, and application layers. This classification facilitated the development of a three-layered architecture, promoting reusability, and consistency in knowledge representation, and serving as the backbone of our semantic framework. Through the utilization of our semantic framework, we accomplished semantic enrichment of both structured and unstructured data. The integration of data from diverse sources involved mapping to ontology concepts, leading to the creation and storage of RDF triples in the triple store. This process resulted in the construction of linked data, ultimately enhancing the discoverability and accessibility of valuable insights. Furthermore, our anomaly detection algorithm effectively leveraged knowledge graphs extracted from the triple store, employing semantic relationships to discern patterns and anomalies within the data. Notably, this capability was exemplified by the identification of correlations between a football game and a COVID-19 event occurring at the same location and time. The framework showcased its capability to address intricate, multi-domain queries and support diverse levels of detail. Additionally, it demonstrated proficiency in data analysis and visualization, generating graphs that depict patterns and trends; however, challenges related to ontology maintenance, alignment, and mapping must be addressed for the approach’s optimal utilization.

Heme, a vital iron-containing molecule, serves fundamental roles in oxygen transport and electron transfer but also acts as an extracellular signaling entity, significantly influencing inflammatory responses. Elevated levels of labile heme resulting from hemolytic events or therapeutic treatments may activate inflammatory signaling pathways, particularly through the Toll-like receptor 4 (TLR4). In this study, we systematically expanded the previously developed Heme Knowledge Graph (HemeKG) to comprehensively incorporate recent findings regarding heme-TLR4 interactions. By employing rigorous literature curation and validation using Biological Expression Language (BEL) standards and the e:BEL Python package, we successfully integrated newly identified molecular entities, notably activator protein 1 (AP-1), interleukin-12 (IL-12), cluster of differentiation 80 (CD80), cluster of differentiation 86 (CD86), and chemokine (C-X-C motif) ligand 1 (CXCL1), into the existing HemeKG framework. Pathway enrichment analysis across Kyoto Encyclopedia of Genes and Genomes (KEGG), Reactome, and WikiPathways databases robustly supported these integrations, consistently identifying significant enrichment of the TLR4 signaling cascade. The updated HemeKG thus provides an integrated and predictive platform, enhancing our understanding of the complex interactions between heme-driven inflammatory pathways and metabolic dysregulation.

MOTIVATION:Knowledge graphs (KGs) in life sciences have become an important application of systems biology as they delineate complex biological and pathophysiological phenomena. They are composed of biological and chemical entities represented with standard ontologies to comply with Findable, Accessible, Interoperable and Reusable (FAIR) principles. Alongside serving as a graph database, KGs hold the potential to address complex scientific queries and facilitate downstream analyses. However, the process of constructing KGs is expensive and time consuming as it primarily relies on manual curation from published literature and experimental data. The existing text-mining workflows are still in their infancy and fail to achieve the accuracy and reliability of manual curation. RESULTS:Knowledge graph generator (KGG) is an automated workflow for representing chemotype and phenotype of diseases and medical conditions. It embeds the underlying schema of curated databases such as OpenTargets, Uniprot, ChEMBL, Integrated Interactions Database and GWAS Central resembling a clockwork-esque mechanism. The resultant KG is a comprehensive and rational assembly of disease-associated entities such as proteins, protein-related pathways, biological processes and functions, genetic variants, chemicals, mechanism of actions, assays and adverse effects. As use cases, we have used KGs to identify shared entities for possible link of comorbidity and compared them with KGs from other sources. We have also demonstrated a use case of identifying putative new targets and repurposing drug candidates in Parkinson's Disease. Lastly, we have developed reusable workflows to explore drug-likeness of chemicals and identify structures of proteins. AVAILABILITY AND IMPLEMENTATION:The resources and codes for KGG are publicly available at: https://github.com/Fraunhofer-ITMP/kgg.

In today's data-centric landscape, effective data stewardship is critical for facilitating scientific research and innovation. This article provides an overview of essential tools and frameworks for modern data stewardship practices. Over 300 tools were analyzed in this study, assessing their utility, relevance to data stewardship, and applicability within the life sciences domain.

MicroRNAs (miRNAs) play important roles in post-transcriptional processes and regulate major cellular functions. The abnormal regulation of expression of miRNAs has been linked to numerous human diseases such as respiratory diseases, cancer, and neurodegenerative diseases. Latest miRNA-disease associations are predominantly found in unstructured biomedical literature. Retrieving these associations manually can be cumbersome and time-consuming due to the continuously expanding number of publications. We propose a deep learning-based text mining approach that extracts normalized miRNA-disease associations from biomedical literature. To train the deep learning models, we build a new training corpus that is extended by distant supervision utilizing multiple external databases. A quantitative evaluation shows that the workflow achieves an area under receiver operator characteristic curve of 98% on a holdout test set for the detection of miRNA-disease associations. We demonstrate the applicability of the approach by extracting new miRNA-disease associations from biomedical literature (PubMed and PubMed Central). We have shown through quantitative analysis and evaluation on three different neurodegenerative diseases that our approach can effectively extract miRNA-disease associations not yet available in public databases. Database URL: https://zenodo.org/records/10523046.

Background: Despite numerous past endeavors for the semantic harmonization of Alzheimer’s disease (AD) cohort studies, an automatic tool has yet to be developed. Objective: As cohort studies form the basis of data-driven analysis, harmonizing them is crucial for cross-cohort analysis. We aimed to accelerate this task by constructing an automatic harmonization tool. Methods: We created a common data model (CDM) through cross-mapping data from 20 cohorts, three CDMs, and ontology terms, which was then used to fine-tune a BioBERT model. Finally, we evaluated the model using three previously unseen cohorts and compared its performance to a string-matching baseline model. Results: Here, we present our AD-Mapper interface for automatic harmonization of AD cohort studies, which outperformed a string-matching baseline on previously unseen cohort studies. We showcase our CDM comprising 1218 unique variables. Conclusion: AD-Mapper leverages semantic similarities in naming conventions across cohorts to improve mapping performance.

Background: Guideline questions are typically proposed by experts. Objective: To assess how large language models (LLMs) can support the development of guideline questions, providing insights on approaches and lessons learned. Design: Two approaches for guideline question generation were assessed: 1) identification of questions conveyed by online search queries and 2) direct generation of guideline questions by LLMs. For the former, the researchers retrieved popular queries on allergic rhinitis using Google Trends (GT) and identified those conveying questions using both manual and LLM-based methods. They then manually structured as guideline questions the queries that conveyed relevant questions. For the second approach, they tasked an LLM with proposing guideline questions, assuming the role of either a patient or a clinician. Setting: Allergic Rhinitis and its Impact on Asthma (ARIA) 2024 guidelines. Participants: None. Measurements: Frequency of relevant questions generated. Results: The authors retrieved 3975 unique queries using GT. From these, they identified 37 questions, of which 22 had not been previously posed by guideline panel members and 2 were eventually prioritized by the panel. Direct interactions with LLMs resulted in the generation of 22 unique relevant questions (11 not previously suggested by panel members), and 4 were eventually prioritized by the panel. In total, 6 of 39 final questions prioritized for the 2024 ARIA guidelines were not initially thought of by the panel. The researchers provide a set of practical insights on the implementation of their approaches based on the lessons learned. Limitation: Single case study (ARIA guidelines). Conclusion: Approaches using LLMs can support the development of guideline questions, complementing traditional methods and potentially augmenting questions prioritized by guideline panels. Primary Funding Source: Fraunhofer Cluster of Excellence for Immune-Mediated Diseases.

Although hundreds of datasets have been published since the beginning of the coronavirus pandemic, there is a lack of centralized resources where these datasets are listed and harmonized to facilitate their applicability and uptake by predictive modeling approaches. Firstly, such a centralized resource provides information about data owners to researchers who are searching datasets to develop their predictive models. Secondly, the harmonization of the datasets supports simultaneously taking advantage of several similar datasets. This, in turn, does not only ease the imperative external validation of data-driven models but can also be used for virtual cohort generation, which helps to overcome data sharing impediments. Here, we present that the COVID-19 data catalogue is a repository that provides a landscape view of COVID-19 studies and datasets as a putative source to enable researchers to develop personalized COVID-19 predictive risk models. The COVID-19 data catalogue currently contains over 400 studies and their relevant information collected from a wide range of global sources such as global initiatives, clinical trial repositories, publications, and data repositories. Further, the curated content stored in this data catalogue is complemented by a web application, providing visualizations of these studies, including their references, relevant information such as measured variables, and the geographical locations of where these studies were performed. This resource is one of the first to capture, organize, and store studies, datasets, and metadata related to COVID-19 in a comprehensive repository. We believe that our work will facilitate future research and development of personalized predictive risk models for COVID-19.

The reconstruction of patient paths—ie, their temporally ordered diagnoses, diagnostic procedures, and the resulting treatments—has shown great potential for gaining new insights in real-world medical practice. Reconstructing patient trajectories from electronic health records (EHRs) is not a new discipline; early attempts go as far back as the late 1990s.1Dionisio JDN Cárdenas AF Taira RK et al.A unified timeline model and user interface for multimedia medical databases.Comput Med Imaging Graph. 1996; 20: 333-346Crossref PubMed Scopus (14) Google Scholar Since then, the field has developed rapidly. Prognosis of symptoms, diagnoses, treatment response, death, or other events has become a field of active research. The application of machine-learning methods and modern artificial intelligence (AI) approaches in particular (eg, transformers such as the Bidirectional Encoder Representations from Transformers [BERT] and generative pretrained transformer [GPT] models) have opened new avenues for modelling patient trajectories;2Lentzen M Linden T Veeranki S et al.A transformer-based model trained on large scale claims data for prediction of severe COVID-19 disease progression.IEEE J Biomed Health Inform. 2023; 27: 4548-4558Crossref Scopus (0) Google Scholar however, the large amount of semantically harmonised yet representative datasets needed for model training imposes a serious obstacle to the generation of high-quality models. Although there is a large body of work regarding time-series modelling and prediction based on structured EHRs (consisting of codes for diagnoses, prescriptions, and diagnostic procedures), the Article presented by Zeljko Kraljevic and colleagues3Kraljevic Z Bean D Shek A et al.Foresight—a generative pretrained transformer for modelling of patient timelines using electronic health records: a retrospective modelling study.Lancet Digit Health. 2024; 6: e281-e290Google Scholar in The Lancet Digital Health reports on a model that uses additional information derived from natural language processing (NLP) annotation of the unstructured (ie, textual) parts of EHRs. In their Article, Kraljevic and colleagues present a GPT approach (Foresight) for modelling patient trajectories using a considerable amount of EHRs as input and making forecasts of the next medical event. They do not only present high prediction performances within a typical split of training and testing data but also by validating model predictions on 34 synthetic patient trajectories by five clinicians. In contrast to previously published work, the authors make extensive use of the unstructured information section within EHRs through a pre-annotation of the EHRs using an established NLP pipeline. Consequently, the GPT is trained on the temporal order of Systematized Nomenclature of Medicine (SNOMED) concepts rather than on the text of the EHRs. The approach covers a wide spectrum of entity types and makes use of a large and diverse set of descriptors for timeline reconstruction and training. Neither the NLP approach, nor the AI modelling strategy used in the Article go beyond what is currently established technology. The training material and the way it has been preprocessed is much more important than the use of a unique technical approach. Others have also used large collections of EHRs for reconstruction of patient trajectories and making forecasts. For example, the authors of the BEHRT,4Li Y Rao S Solares JRA et al.BEHRT: transformer for electronic health records.Sci Rep. 2020; 107155 Google Scholar G-BERT,5Shang J, Ma T, Xiao C, Sun J. Pre-training of graph augmented transformers for medication recommendation. Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence AI for Improving Human Well-being; Aug 10–16, 2019 (pp 5953–59).Google Scholar CEHR-BERT,6Pang C Jiang X Kalluri KS et al.CEHR-BERT: Incorporating temporal information from structured EHR data to improve prediction tasks.Proc Mach Learn Health. 2021; 158: 239-260Google Scholar and Med-BERT7Rasmy L Xiang Y Xie Z Tao C Zhi D Med-BERT: pretrained contextualized embeddings on large-scale structured electronic health records for disease prediction.NPJ Digit Med. 2021; 4: 86Crossref PubMed Scopus (217) Google Scholar models have used millions of structured EHRs. The Article discussed here must be seen as part of a growing number of related publications that use slightly different input sources for patient trajectory prediction. We will hopefully see lots of such time-series forecasting models from different health-care systems and different regions of the world, resulting in a diverse selection of generative models that allow for the representation of a wide spectrum of individual patient journeys. However, the main questions remain: what does the model tell us now? What can we learn? Kraljevic and colleagues refrain from making bold claims and present Foresight as model with potential, both clinical and educational. But what can we learn beyond the use of Foresight in the training of medical professionals? Although there are always time and resource constraints in the development of predictive models, there are some limitations to Kraljevic and colleagues' approach. It would have been desirable to see the trajectory reconstruction and prediction engine challenged by the reality of lifelong trajectories that, for instance, Siggaard and colleagues have established for the Danish population.8Siggaard T Reguant R Jørgensen IF et al.Disease trajectory browser for exploring temporal, population-wide disease progression patterns in 7.2 million Danish patients.Nat Commun. 2020; 114952 Crossref PubMed Scopus (53) Google Scholar That would have been a truly relevant validation, and it would have—at least partially—addressed the generalisability aspect of the model. Another obvious validation scenario would have been the comparison of best practice according to clinical guidelines with the reality of the patient paths that are encoded in the model.9Canonica GW Agache I Schünemann HJ Roche N Price D Del Giacco S Next generation health guidelines: the role of real-life data in evidence-based medicine.Allergy. 2024; 79: 12-14Crossref Scopus (0) Google Scholar Nonetheless, we should be very pleased with the work of Kraljevic and colleagues. If more models such as these are developed and validated, we might develop a larger ecosystem of patient timelines based on many models and will have plenty of material to address the above limitations. We declare no competing interests. Foresight—a generative pretrained transformer for modelling of patient timelines using electronic health records: a retrospective modelling studyForesight is a general-purpose model for biomedical concept modelling that can be used for real-world risk forecasting, virtual trials, and clinical research to study the progression of disorders, to simulate interventions and counterfactuals, and for educational purposes. Full-Text PDF Open Access