Large Language Models (LLMs) excel at language understanding but remain limited in knowledge-intensive domains due to hallucinations, outdated information, and limited explainability. Text-based retrieval-augmented generation (RAG) helps ground model outputs in external sources but struggles with multi-hop reasoning. Knowledge Graphs (KGs), in contrast, support precise, explainable querying, yet require a knowledge of query languages. This work introduces an interactive framework in which LLMs generate and explain Cypher graph queries and users iteratively refine them through natural language. Applied to real-world KGs, the framework improves accessibility to complex datasets while preserving factual accuracy and semantic rigor and provides insight into how model performance varies across domains. Our core quantitative evaluation is a 90-query benchmark on a synthetic movie KG that measures query explanation quality and fault detection across multiple LLMs, complemented by two smaller real-life query-generation experiments on a Hyena KG and the MaRDI (Mathematical Research Data Initiative) KG.

The evolution of many dynamical systems that describe relationships or interactions between objects can be effectively modeled by temporal networks, which are typically represented as a sequence of static network snapshots. In this paper, we introduce a novel random walk-based approach that can identify clusters of time-snapshots in which network community structures are stable. This allows us to detect significant structural shifts over time, such as the splitting or merging of communities or their births and deaths. We also provide a low-dimensional representation of entire snapshots, placing those with similar community structure close to each other in the feature space. To validate our approach, we develop an agent-based algorithm that generates synthetic datasets with the desired characteristic properties, enabling thorough testing and benchmarking. We further demonstrate the effectiveness and broad applicability of our technique by testing it on various social dynamics models and real-world datasets and comparing its performance to several state-of-the-art algorithms. Our findings highlight the strength of our approach to correctly capture and analyze the dynamics of complex systems.

This paper introduces a novel hybrid model combining Partial Differential Equations (PDEs) and Ordinary Differential Equations (ODEs) to simulate infectious disease dynamics across geographic regions. By leveraging the spatial detail of PDEs and the computational efficiency of ODEs, the model enables rapid evaluation of public health interventions. Applied to synthetic environments and real-world scenarios in Lombardy, Italy, and Berlin, Germany, the model highlights how interactions between PDE and ODE regions affect infection dynamics, especially in high-density areas. Key findings reveal that the placement of model boundaries in densely populated regions can lead to inaccuracies in infection spread, suggesting that boundaries should be positioned in areas of lower population density to better reflect transmission dynamics. Additionally, regions with low population density hinder infection flow, indicating a need for incorporating, e.g., jumps in the model to enhance its predictive capabilities. Results indicate that the hybrid model achieves a balance between computational speed and accuracy, making it a valuable tool for policymakers in real-time decision-making and scenario analysis in epidemiology and potentially in other fields requiring similar modeling approaches.

Accurate diagnosis of heart arrhythmias requires the interpretation of electrocardiograms (ECG), which capture the electrical activity of the heart. Automating this process through machine learning is challenging due to the need for large annotated datasets, which are difficult and costly to collect. To address this issue, transfer learning is often employed, where models are pre-trained on large datasets and fine-tuned for specific ECG classification tasks with limited labeled data. Self-supervised learning has become a widely adopted pre-training method, enabling models to learn meaningful representations from unlabeled datasets. In this work, we explore the joint-embedding predictive architecture (JEPA) for self-supervised learning from ECG data. Unlike invariance-based methods, JEPA does not rely on hand-crafted data augmentations, and unlike generative methods, it predicts latent features rather than reconstructing input data. We create a large unsupervised pre-training dataset by combining ten public ECG databases, amounting to over one million records. We pre-train Vision Transformers using JEPA on this dataset and fine-tune them on various PTB-XL benchmarks. Our results show that JEPA outperforms existing invariance-based and generative approaches, achieving an AUC of 0.945 on the PTB-XL all statements task. JEPA consistently learns the highest quality representations, as demonstrated in frozen evaluations, and proves advantageous for pre-training even in the absence of additional data.

This study investigates the spatial integration of agent-based models (ABMs) and compartmental models for infectious disease modeling, presenting a novel hybrid approach and examining its implications. ABMs offer detailed insights by simulating interactions and decisions among individuals but are computationally expensive for large populations. Compartmental models capture population-level dynamics more efficiently but lack granular detail. We developed a hybrid model that aims to balance the granularity of ABMs with the computational efficiency of compartmental models, offering a more nuanced understanding of disease spread in diverse scenarios, including large populations. This model spatially couples discrete and continuous populations by integrating an ordinary differential equation model with a spatially explicit ABM. Our key objectives were to systematically assess the consistency of disease dynamics and the computational efficiency across various configurations. For this, we evaluated two experimental scenarios and varied the influence of each sub-model via spatial distribution. In the first, the ABM component modeled a homogeneous population; in the second, it simulated a heterogeneous population with landscape-driven movement. Results show that the hybrid model can significantly reduce computational costs but is sensitive to between-model differences, highlighting the importance of model equivalence in hybrid approaches. The code is available at: git.zib.de/ibostanc/hybrid_abm_ode

Recommender systems suffer from the cold-start problem whenever a new user joins the platform or a new item is added to the catalog. To address item cold-start, we propose to replace the embedding layer in sequential recommenders with a dynamic storage that has no learnable weights and can keep an arbitrary number of representations. In this paper, we present FELRec, a large embedding network that refines the existing representations of users and items in a recursive manner, as new information becomes available. In contrast to similar approaches, our model represents new users and items without side information or time-consuming fine-tuning. During item cold-start, our method outperforms similar method by 29.50%-47.45%. Further, our proposed model generalizes well to previously unseen datasets. The source code is publicly available at github.com/kweimann/FELRec.

Nirmatrelvir/Ritonavir is an oral treatment for mild to moderate COVID-19 cases with a high risk for a severe course of the disease. For this paper, a comprehensive literature review was performed, leading to a summary of currently available data on Nirmatrelvir/Ritonavir’s ability to reduce the risk of progressing to a severe disease state. Herein, the focus lies on publications that include comparisons between patients receiving Nirmatrelvir/Ritonavir and a control group. The findings can be summarized as follows: Data from the time when the Delta-variant was dominant show that Nirmatrelvir/Ritonavir reduced the risk of hospitalization or death by 88.9% for unvaccinated, non-hospitalized high-risk individuals. Data from the time when the Omicron variant was dominant found decreased relative risk reductions for various vaccination statuses: between 26% and 65% for hospitalization. The presented papers that differentiate between unvaccinated and vaccinated individuals agree that unvaccinated patients benefit more from treatment with Nirmatrelvir/Ritonavir. However, when it comes to the dependency of potential on age and comorbidities, further studies are necessary. From the available data, one can conclude that Nirmatrelvir/Ritonavir cannot substitute vaccinations; however, its low manufacturing cost and easy administration make it a valuable tool in fighting COVID-19, especially for countries with low vaccination rates.

Recent advances in deep learning and natural language generation have significantly improved image captioning, enabling automated, human-like descriptions for visual content. In this work, we apply these captioning techniques to generate clinician-like interpretations of ECG data. This study leverages existing ECG datasets accompanied by free-text reports authored by healthcare professionals (HCPs) as training data. These reports, while often inconsistent, provide a valuable foundation for automated learning. We introduce an encoder-decoder-based method that uses these reports to train models to generate detailed descriptions of ECG episodes. This represents a significant advancement in ECG analysis automation, with potential applications in zero-shot classification and automated clinical decision support. The model is tested on various datasets, including both 1- and 12-lead ECGs. It significantly outperforms the state-of-the-art reference model by Qiu et al., achieving a METEOR score of 55.53 reference model. Furthermore, several key design choices are discussed, providing a comprehensive overview of current challenges and innovations in this domain. The source codes for this research are publicly available in our Git repository https://git.zib.de/ableich/ecg-comment-generation-public

Advancements in natural language processing have revolutionized the way we can interact with digital information systems, such as databases, making them more accessible. However, challenges persist, especially when accuracy is critical, as in the biomedical domain. A key issue is the hallucination problem, where models generate information unsupported by the underlying data, potentially leading to dangerous misinformation. This paper presents a novel approach designed to bridge this gap by combining Large Language Models (LLM) and Knowledge Graphs (KG) to improve the accuracy and reliability of question-answering systems, on the example of a biomedical KG. Built on the LangChain framework, our method incorporates a query checker that ensures the syntactical and semantic validity of LLM-generated queries, which are then used to extract information from a Knowledge Graph, substantially reducing errors like hallucinations. We evaluated the overall performance using a new benchmark dataset of 50 biomedical questions, testing several LLMs, including GPT-4 Turbo and llama3:70b. Our results indicate that while GPT-4 Turbo outperforms other models in generating accurate queries, open-source models like llama3:70b show promise with appropriate prompt engineering. To make this approach accessible, a user-friendly web-based interface has been developed, allowing users to input natural language queries, view generated and corrected Cypher queries, and verify the resulting paths for accuracy. Overall, this hybrid approach effectively addresses common issues such as data gaps and hallucinations, offering a reliable and intuitive solution for question answering systems. The source code for generating the results of this paper and for the user-interface can be found in our Git repository: https://git.zib.de/lpusch/cyphergenkg-gui

Respiratory viral infections (RVIs) are common reasons for healthcare consultations. The inpatient management of RVIs consumes significant resources. From 2009 to 2014, we assessed the costs of RVI management in 4776 hospitalized children aged 0–18 years participating in a quality improvement program, where all ILI patients underwent virologic testing at the National Reference Centre followed by detailed recording of their clinical course. The direct (medical or non-medical) and indirect costs of inpatient management outside the ICU (‘non-ICU’) versus management requiring ICU care (‘ICU’) added up to EUR 2767.14 (non-ICU) vs. EUR 29,941.71 (ICU) for influenza, EUR 2713.14 (non-ICU) vs. EUR 16,951.06 (ICU) for RSV infections, and EUR 2767.33 (non-ICU) vs. EUR 14,394.02 (ICU) for human rhinovirus (hRV) infections, respectively. Non-ICU inpatient costs were similar for all eight RVIs studied: influenza, RSV, hRV, adenovirus (hAdV), metapneumovirus (hMPV), parainfluenza virus (hPIV), bocavirus (hBoV), and seasonal coronavirus (hCoV) infections. ICU costs for influenza, however, exceeded all other RVIs. At the time of the study, influenza was the only RVI with antiviral treatment options available for children, but only 9.8% of influenza patients (non-ICU) and 1.5% of ICU patients with influenza received antivirals; only 2.9% were vaccinated. Future studies should investigate the economic impact of treatment and prevention of influenza, COVID-19, and RSV post vaccine introduction.

The sharing and citation of research data is becoming increasingly recognized as an essential building block in scientific research across various fields and disciplines. Sharing research data allows other researchers to reproduce results, replicate findings, and build on them. Ultimately, this will foster faster cycles in knowledge generation. Some disciplines, such as astronomy or bioinformatics, already have a long history of sharing data; many others do not. The current landscape of available systems for sharing research data is diverse. In this article, we conduct a detailed analysis of existing web-based systems, specifically focusing on mathematical research data.

Background: Collaborative comparisons and combinations of epidemic models are used as policy-relevant evidence during epidemic outbreaks. In the process of collecting multiple model projections, such collaborations may gain or lose relevant information. Typically, modellers contribute a probabilistic summary at each time-step. We compared this to directly collecting simulated trajectories. We aimed to explore information on key epidemic quantities; ensemble uncertainty; and performance against data, investigating potential to continuously gain information from a single cross-sectional collection of model results. Methods: We compared projections from the European COVID-19 Scenario Modelling Hub. Five teams modelled incidence in Belgium, the Netherlands, and Spain. We compared July 2022 projections by incidence, peaks, and cumulative totals. We created a probabilistic ensemble drawn from all trajectories, and compared to ensembles from a median across each model’s quantiles, or a linear opinion pool. We measured the predictive accuracy of individual trajectories against observations, using this in a weighted ensemble. We repeated this sequentially against increasing weeks of observed data. We evaluated these ensembles to reflect performance with varying observed data. Results: By collecting modelled trajectories, we showed policy-relevant epidemic characteristics. Trajectories contained a right-skewed distribution well represented by an ensemble of trajectories or a linear opinion pool, but not models’ quantile intervals. Ensembles weighted by performance typically retained the range of plausible incidence over time, and in some cases narrowed this by excluding some epidemic shapes. Conclusions: We observed several information gains from collecting modelled trajectories rather than quantile distributions, including potential for continuously updated information from a single model collection. The value of information gains and losses may vary with each collaborative effort’s aims, depending on the needs of projection users. Understanding the differing information potential of methods to collect model projections can support the accuracy, sustainability, and communication of collaborative infectious disease modelling efforts.

Table processing, a key task in natural language processing, has significantly benefited from recent advancements in language models (LMs). However, the capabilities of LMs in table-to-text generation, which transforms structured data into coherent narrative text, require an in-depth investigation, especially with current open-source models. This study explores the effectiveness of various in-context learning strategies in LMs across benchmark datasets, focusing on the impact of providing examples to the model. More importantly, we examine a real-world use case, offering valuable insights into practical applications. To complement traditional evaluation metrics, we employ a large language model (LLM) self-evaluation approach using chain-of-thought reasoning and assess its correlation with human-aligned metrics like BERTScore. Our findings highlight the significant impact of examples in improving table-to-text generation and suggest that, while LLM self-evaluation has potential, its current alignment with human judgment could be enhanced. This points to the need for more reliable evaluation methods.

In response to increasing data privacy regulations, this work examines the use of federated learning for deep residual networks to diagnose cardiac abnormalities from electrocardiogram (ECG) data. This approach allows medical institutions to collaborate without exchanging raw patient data. We utilize the publicly available data from the PhysioNet/Computing in Cardiology Challenge 2021, featuring diverse ECG databases, to compare the classification performance of three federated learning methods against both central training with data sharing and isolated training scenarios. We show that federated learning outperforms ECG classifiers trained in isolation. In particular, our findings demonstrate that a globally trained model fine-tuned to specific local datasets surpasses non-collaborative approaches. This shows that models trained in federation learn general features that can be tailored to specific tasks. Furthermore, federated learning almost matches the performance of central training with data sharing on out-of-distribution data from non-participating institutions. These results highlight the ability of federated learning in developing models that generalize well across diverse patient data, without the need to share data among institutions, thus addressing data privacy concerns.

The recent COVID-19 pandemic has thrown the importance of accurately forecasting contagion dynamics and learning infection parameters into sharp focus. At the same time, effective policy-making requires knowledge of the uncertainty on such predictions, in order, for instance, to be able to ready hospitals and intensive care units for a worst-case scenario without needlessly wasting resources. In this work, we apply a novel and powerful computational method to the problem of learning probability densities on contagion parameters and providing uncertainty quantification for pandemic projections. Using a neural network, we calibrate an ODE model to data of the spread of COVID-19 in Berlin in 2020, achieving both a significantly more accurate calibration and prediction than Markov-Chain Monte Carlo (MCMC)-based sampling schemes. The uncertainties on our predictions provide meaningful confidence intervals e.g. on infection figures and hospitalisation rates, while training and running the neural scheme takes minutes where MCMC takes hours. We show convergence of our method to the true posterior on a simplified SIR model of epidemics, and also demonstrate our method's learning capabilities on a reduced dataset, where a complex model is learned from a small number of compartments for which data is available.

Agent-based models (ABMs) offer a powerful framework for understanding complex systems. However, their computational demands often become a significant barrier as the number of agents and complexity of the simulation increase. Traditional ABM platforms often struggle to fully exploit modern computing resources, hindering the development of large-scale simulations. This paper presents Vahana.jl, a high performance computing open source framework that aims to address these limitations. Building on the formalism of synchronous graph dynamical systems, Vahana.jl is especially well suited for models with a focus on (social) networks. The framework seamlessly supports distribution across multiple compute nodes, enabling simulations that would otherwise be beyond the capabilities of a single machine. Implemented in Julia, Vahana.jl leverages the interactive Read-Eval-Print Loop (REPL) environment, facilitating rapid model development and experimentation.

Measurable levels of immunoglobulin G antibodies develop after infections with and vaccinations against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). These antibody levels are dynamic: due to waning, antibody levels will drop over time. During the COVID-19 pandemic, multiple models predicting infection dynamics were used by policymakers to support the planning of public health policies. Explicitly integrating antibody and waning effects into the models is crucial for reliable calculations of individual infection risk. However, only few approaches have been suggested that explicitly treat these effects. This paper presents a methodology that explicitly models antibody levels and the resulting protection against infection for individuals within an agent-based model. The model was developed in response to the complexity of different immunization sequences and types and is based on neutralization titer studies. This approach allows complex population studies with explicit antibody and waning effects. We demonstrate the usefulness of our model in two use cases.

Aims: Interpretation of continuous EEG is a demanding task that requires the expertise of trained neurologists. However, these experts are not always available in many medical centers. As part of the 2023 George B. Moody PhysioNet Challenge, we developed a deep learning based method for analyzing EEG data of comatose patients and predicting prognosis following cardiac arrest. Methods: Our approach is a two-step pipeline that consists of a prediction model and a decision-making strategy. The prediction model is a residual neural network (ResNet-18) that extracts features and makes a prediction based on a short 5-minute EEG recording. In the second step, a majority vote over multiple predictions made for several EEG recordings of a patient determines the final prognosis. Results: Based on 10-fold cross-validation on the training set, we achieved a true positive rate (TPR) of 0.41 for predicting poor outcome while keeping the false positive rate below 0.05 at 72 hours after recovery of spontaneous circulation. On the official challenge leaderboard, our team ZIB_Visual scored 0.426 TPR. Conclusion: Our approach, while simple to implement and execute, faced overfitting challenges during the official competition phase. In this paper, we discuss our implementation and potential improvements to address these issues.

Background Collaborative comparisons and combinations of epidemic models are used as policy-relevant evidence during epidemic outbreaks. In the process of collecting multiple model projections, such collaborations may gain or lose relevant information. Typically, modellers contribute a probabilistic summary at each time-step. We compared this to directly collecting simulated trajectories. We aimed to explore information on key epidemic quantities; ensemble uncertainty; and performance against data, investigating potential to continuously gain information from a single cross-sectional collection of model results. Methods We compared July 2022 projections from the European COVID-19 Scenario Modelling Hub. Five modelling teams projected incidence in Belgium, the Netherlands, and Spain. We compared projections by incidence, peaks, and cumulative totals. We created a probabilistic ensemble drawn from all trajectories, and compared to ensembles from a median across each model’s quantiles, or a linear opinion pool. We measured the predictive accuracy of individual trajectories against observations, using this in a weighted ensemble. We repeated this sequentially against increasing weeks of observed data. We evaluated these ensembles to reflect performance with varying observed data. Results By collecting modelled trajectories, we showed policy-relevant epidemic characteristics. Trajectories contained a right-skewed distribution well represented by an ensemble of trajectories or a linear opinion pool, but not models’ quantile intervals. Ensembles weighted by performance typically retained the range of plausible incidence over time, and in some cases narrowed this by excluding some epidemic shapes. Conclusions We observed several information gains from collecting modelled trajectories rather than quantile distributions, including potential for continuously updated information from a single model collection. The value of information gains and losses may vary with each collaborative effort’s aims, depending on the needs of projection users. Understanding the differing information potential of methods to collect model projections can support the accuracy, sustainability, and communication of collaborative infectious disease modelling efforts. Data availability All code and data available on Github: https://github.com/covid19-forecast-hub-europe/aggregation-info-loss

Supplementary Data from Serum Peptidome Profiling Revealed Platelet Factor 4 as a Potential Discriminating Peptide Associated with Pancreatic Cancer