Journal of Biomedical Informatics最新文献

{"title":"GutGPT: A multidimensional knowledge-enhanced large language model for gastrointestinal medicine","authors":"Rui-Ya Zhang ,&nbsp;Peng-Peng Qiang ,&nbsp;Yu-Xia Hao ,&nbsp;Hong-Ye Tan ,&nbsp;Kai Zhao ,&nbsp;Ling-Jun Cai ,&nbsp;Jun-Ping Wang","doi":"10.1016/j.jbi.2025.104885","DOIUrl":"10.1016/j.jbi.2025.104885","url":null,"abstract":"<div><h3>Background</h3><div>Gastrointestinal (GI) diseases are common, chronic conditions that require personalized, long-term management, placing a heavy burden on traditional healthcare systems. While large language models (LLMs) offer potential for supporting patient care with personalized and empathetic guidance, existing models often lack domain-specific knowledge in GI diseases and suffer from issues like slow convergence and overfitting.</div></div><div><h3>Methodology</h3><div>We first construct a high-quality GI disease QA dataset comprising 191,615 entries from diverse sources: real-world doctor-patient dialogues, medical knowledge graphs, medical guidelines, and Chinese medical licensing exam data. Then, we introduce GutGPT, an LLM fine-tuned from Baichuan-13B-Chat using Low-Rank Adaptation (LoRA) technology with self-attention mechanism parameter sharing. To evaluate the performance of GutGPT and other existing LLMs, we use a combination of expert evaluation and public dataset testing to comprehensively assess each model’s accuracy and empathy.</div></div><div><h3>Results</h3><div>We conduct comprehensive evaluations, including expert evaluations and evaluations on multiple benchmark datasets. The results show that our model outperforms 16 existing methods and achieves state-of-the-art performance. In expert evaluations, GutGPT improves diagnostic accuracy by 9.59% compared to the baselines. On two public medical QA datasets, CMB and CMExam, it achieves an average accuracy improvement of 22.47%.</div></div><div><h3>Conclusions</h3><div>GutGPT achieves high accuracy in managing GI disease patients and demonstrates strong empathy. It serves as an important auxiliary tool for both patients and physicians in disease management.</div></div>","PeriodicalId":15263,"journal":{"name":"Journal of Biomedical Informatics","volume":"169 ","pages":"Article 104885"},"PeriodicalIF":4.5,"publicationDate":"2025-07-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144721894","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Adaptive-AutoMO: A domain adaptive automated multiobjective neural network for reliable lesion malignancy prediction via digital breast tomosynthesis","authors":"Xi Chen ,&nbsp;Jiahuan Lv ,&nbsp;Zeyu Wang ,&nbsp;Genggeng Qin ,&nbsp;Zhiguo Zhou","doi":"10.1016/j.jbi.2025.104869","DOIUrl":"10.1016/j.jbi.2025.104869","url":null,"abstract":"<div><div>Early diagnosis of breast cancer remains a significant global health challenge, and the potential use of deep learning in Digital Breast Tomosynthesis (DBT) based breast cancer diagnosis is a promising avenue. To address data scarcity and domain shift problems in building a lesion malignancy predictive model, we proposed a domain adaptive automated multiobjective neural network (Adaptive-AutoMO) for reliable lesion malignancy prediction via DBT. Adaptive-AutoMO addresses three key challenges simultaneously, they are: privacy preserving, credibility measurement, and balance, which consists of training, adaptation and testing stages. In the training stage, we developed a multiobjective immune neural architecture search algorithm (MINAS) to generate a Pareto-optimal model set with balanced sensitivity and specificity and introduced a Bayesian optimization algorithm to optimize the hyperparameters. In the adaptation stage, a semi-supervised domain adaptive feature network based on maximum mean discrepancy (MMD-SSDAF) was designed, which can make the balanced models adaptable to the target domain and preserve the data privacy in the source domain. In the testing stage, we proposed an evidence reasoning method based on entropy (ERE) that can fuse multiple adapted models and estimate uncertainty to improve the model credibility. The experiments on two DBT image datasets (source and target domain datasets) revealed that Adaptive-AutoMO outperformed ResNet-18, DenseNet-121, and other available domain adaptive models. Meanwhile, the removal of high uncertainty samples resulted in a performance improvement in the target domain. These experiments affirmed that Adaptive-AutoMO can not only enhance model’s performance, but also preserve privacy in the source domain data, boost model credibility, and achieve a balance between sensitivity and specificity.</div></div>","PeriodicalId":15263,"journal":{"name":"Journal of Biomedical Informatics","volume":"169 ","pages":"Article 104869"},"PeriodicalIF":4.5,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144717986","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A review on knowledge graphs for healthcare: Resources, applications, and promises","authors":"Hejie Cui ,&nbsp;Jiaying Lu ,&nbsp;Ran Xu ,&nbsp;Shiyu Wang ,&nbsp;Wenjing Ma ,&nbsp;Yue Yu ,&nbsp;Shaojun Yu ,&nbsp;Xuan Kan ,&nbsp;Chen Ling ,&nbsp;Liang Zhao ,&nbsp;Zhaohui S. Qin ,&nbsp;Joyce C. Ho ,&nbsp;Tianfan Fu ,&nbsp;Jing Ma ,&nbsp;Mengdi Huai ,&nbsp;Fei Wang ,&nbsp;Carl Yang","doi":"10.1016/j.jbi.2025.104861","DOIUrl":"10.1016/j.jbi.2025.104861","url":null,"abstract":"<div><h3>Objective:</h3><div>This comprehensive review aims to provide an overview of the current state of Healthcare Knowledge Graphs (HKGs), including their construction, utilization models, and applications across various healthcare and biomedical research domains.</div></div><div><h3>Methods:</h3><div>We thoroughly analyzed existing literature on HKGs, covering their construction methodologies, utilization techniques, and applications in basic science research, pharmaceutical research and development, clinical decision support, and public health. The review encompasses both model-free and model-based utilization approaches and the integration of HKGs with large language models (LLMs).</div></div><div><h3>Results:</h3><div>We searched Google Scholar for relevant papers on HKGs and classified them into the following topics: HKG construction, HKG utilization, and their downstream applications in various domains. We also discussed their special challenges and the promise for future work.</div></div><div><h3>Discussion:</h3><div>The review highlights the potential of HKGs to significantly impact biomedical research and clinical practice by integrating vast amounts of biomedical knowledge from multiple domains. The synergy between HKGs and LLMs offers promising opportunities for constructing more comprehensive knowledge graphs and improving the accuracy of healthcare applications.</div></div><div><h3>Conclusions:</h3><div>HKGs have emerged as a powerful tool for structuring medical knowledge, with broad applications across biomedical research, clinical decision-making, and public health. This survey serves as a roadmap for future research and development in the field of HKGs, highlighting the potential of combining knowledge graphs with advanced machine learning models for healthcare transformation.</div></div>","PeriodicalId":15263,"journal":{"name":"Journal of Biomedical Informatics","volume":"169 ","pages":"Article 104861"},"PeriodicalIF":4.5,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144717985","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Federated causal discovery with missing data in a multicentric study on endometrial cancer","authors":"Alessio Zanga ,&nbsp;Alice Bernasconi ,&nbsp;Peter J.F. Lucas ,&nbsp;Hanny Pijnenborg ,&nbsp;Casper Reijnen ,&nbsp;Marco Scutari ,&nbsp;Anthony C. Constantinou","doi":"10.1016/j.jbi.2025.104877","DOIUrl":"10.1016/j.jbi.2025.104877","url":null,"abstract":"<div><h3>Objectives:</h3><div>Establishing causal dependencies is crucial in applied domains, such as medicine and healthcare, where decision-making must be explainable. In these settings, small sample sizes and missing data call for federated approaches to maximise the amount of information we can use.</div></div><div><h3>Methods:</h3><div>We propose a novel federated causal discovery algorithm capable of pooling information from multiple sources with heterogeneous missing data to learn a graph representing cause–effect relationships. In particular, we learn a causal graph on a centralised server while taking into account both prior knowledge and missingness mechanism specific to each client.</div></div><div><h3>Results:</h3><div>We applied the proposed algorithm to synthetic data and real-world data from a multicentric study on endometrial cancer, validating the obtained causal graph through quantitative analyses and a clinical literature review.</div></div><div><h3>Conclusion:</h3><div>Our approach learns an accurate model despite data missing not-at-random.</div></div>","PeriodicalId":15263,"journal":{"name":"Journal of Biomedical Informatics","volume":"169 ","pages":"Article 104877"},"PeriodicalIF":4.5,"publicationDate":"2025-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144707575","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A scoping review of natural language processing in addressing medically inaccurate information: Errors, misinformation, and hallucination","authors":"Zhaoyi Sun ,&nbsp;Wen-Wai Yim ,&nbsp;Özlem Uzuner ,&nbsp;Fei Xia ,&nbsp;Meliha Yetisgen","doi":"10.1016/j.jbi.2025.104866","DOIUrl":"10.1016/j.jbi.2025.104866","url":null,"abstract":"<div><h3>Objective:</h3><div>This review aims to explore the potential and challenges of using Natural Language Processing (NLP) to detect, correct, and mitigate medically inaccurate information, including errors, misinformation, and hallucination. By unifying these concepts, the review emphasizes their shared methodological foundations and their distinct implications for healthcare. Our goal is to advance patient safety, improve public health communication, and support the development of more reliable and transparent NLP applications in healthcare.</div></div><div><h3>Methods:</h3><div>A scoping review was conducted following PRISMA-ScR guidelines, analyzing studies from 2020 to 2024 across five databases. Studies were selected based on their use of NLP to address medically inaccurate information and were categorized by topic, tasks, document types, datasets, models, and evaluation metrics.</div></div><div><h3>Results:</h3><div>NLP has shown potential in addressing medically inaccurate information on the following tasks: (1) error detection (2) error correction (3) misinformation detection (4) misinformation correction (5) hallucination detection (6) hallucination mitigation. However, challenges remain with data privacy, context dependency, and evaluation standards.</div></div><div><h3>Conclusion:</h3><div>This review highlights the advancements in applying NLP to tackle medically inaccurate information while underscoring the need to address persistent challenges. Future efforts should focus on developing real-world datasets, refining contextual methods, and improving hallucination management to ensure reliable and transparent healthcare applications.</div></div>","PeriodicalId":15263,"journal":{"name":"Journal of Biomedical Informatics","volume":"169 ","pages":"Article 104866"},"PeriodicalIF":4.0,"publicationDate":"2025-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144686639","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Variational temporal deconfounder network for individualized treatment effect estimation with longitudinal observational data","authors":"Hao Dai ,&nbsp;Yu Huang ,&nbsp;Yuxi Liu ,&nbsp;Xing He ,&nbsp;Jingchuan Guo ,&nbsp;Mattia Prosperi ,&nbsp;Jiang Bian","doi":"10.1016/j.jbi.2025.104880","DOIUrl":"10.1016/j.jbi.2025.104880","url":null,"abstract":"<div><h3>Objective</h3><div>By leveraging real-world electronic health record (EHR) data, this study set out to estimate individualized treatment effects (ITE) in longitudinal observational settings to advance personalized medicine, addressing key challenges that are often observed in real-world clinical scenarios and pose statistical challenges, including hidden confounding and dynamic treatment regimens.</div></div><div><h3>Methods</h3><div>We propose the Variational Temporal Deconfounder Network (VTDNet), a novel framework designed to account for time-varying hidden confounding using a variational recurrent transformer-based autoencoder. Specifically, VTDNet comprises three critical components: a temporal Encoder-Decoder structure to capture hidden representation, a Treatment Block that captures interdependencies among multiple treatments, and a Potential Outcome Block that predicts both factual and counterfactual outcomes. We assess the effectiveness of the proposed framework using a synthetic dataset and two real-world datasets: MIMIC-III, an EHR dataset focusing on intensive care settings, and NACC, emphasizing neurodegenerative disease, collected using a standardized protocol from participants enrolled in Alzheimer’s Disease Research Center (ADRC) clinical cores.</div></div><div><h3>Results</h3><div>Experimental results on the synthetic dataset demonstrate superior accuracy under varying levels of confounding. On real-world EHR datasets, VTDNet achieves lower root mean squared error, mean absolute error, and influence function precision in the estimation of heterogeneous effects compared to existing state-of-the-art methods.</div></div><div><h3>Conclusion</h3><div>The proposed VTDNet offers a robust framework for estimating individualized treatment effects in longitudinal settings, effectively accommodating irregular time points and high-dimensional data while addressing hidden confounders through a deep generative approach. It holds significant potential to advance personalized medicine and support real-world evidence generation. Future work will aim to extend VTDNet to continuous treatment scenarios, such as dose–response analysis, to further broaden its applicability in clinical practice.</div></div>","PeriodicalId":15263,"journal":{"name":"Journal of Biomedical Informatics","volume":"169 ","pages":"Article 104880"},"PeriodicalIF":4.0,"publicationDate":"2025-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144695011","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Natural language processing for scalable feature engineering and ultra-high-dimensional confounding adjustment in healthcare database studies","authors":"Richard Wyss ,&nbsp;Jie Yang ,&nbsp;Sebastian Schneeweiss ,&nbsp;Joseph M. Plasek ,&nbsp;Li Zhou ,&nbsp;Thomas Deramus ,&nbsp;Janick G. Weberpals ,&nbsp;Kerry Ngan ,&nbsp;Theodore N. Tsacogianis ,&nbsp;Kueiyu Joshua Lin","doi":"10.1016/j.jbi.2025.104882","DOIUrl":"10.1016/j.jbi.2025.104882","url":null,"abstract":"<div><h3>Background</h3><div>To improve confounding control in healthcare database studies, data-driven algorithms may empirically identify and adjust for large numbers of pre-exposure variables that indirectly capture information on unmeasured confounding factors (‘proxy’ confounders). Current approaches for high-dimensional proxy adjustment do not leverage free-text notes from electronic health records (EHRs). Unsupervised natural language processing (NLP) technology can scale to generate large numbers of structured features from unstructured notes.</div></div><div><h3>Objective</h3><div>To assess the impact of supplementing claims data analyses with large numbers of NLP generated features for high-dimensional proxy adjustment.</div></div><div><h3>Methods</h3><div>We linked Medicare claims with EHR data to generate three cohorts comparing different classes of medications on the 6-month risk of cardiovascular outcomes. We used various NLP methods to generate structured features from free-text EHR notes and used least absolute shrinkage and selection operator (LASSO) regression to fit several propensity score (PS) models that included different covariate sets as candidate predictors. Covariate sets included features generated from claims data only, and claims data plus NLP-generated EHR features.</div></div><div><h3>Results</h3><div>Including both claims codes and NLP-generated EHR features as candidate predictors improved overall covariate balance with standardized differences being &lt; 0.1 for all variables. While overall balance improved, the impact on estimated treatment effects was more nuanced with adjustment for NLP-generated features moving effect estimates further in the expected direction in two of the empirical studies but had no impact on the third study.</div></div><div><h3>Conclusion</h3><div>Supplementing administrative claims with large numbers of NLP-generated features for ultra-high-dimensional proxy confounder adjustment improved overall covariate balance and may provide a modest benefit in terms of capturing confounder information.</div></div>","PeriodicalId":15263,"journal":{"name":"Journal of Biomedical Informatics","volume":"169 ","pages":"Article 104882"},"PeriodicalIF":4.0,"publicationDate":"2025-07-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144682653","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A multifaceted approach to drug–drug interaction extraction with fusion strategies","authors":"Ba-Hoang Tran ,&nbsp;Hung-Manh Hoang ,&nbsp;Binh-Nguyen Nguyen ,&nbsp;Duy-Cat Can ,&nbsp;Hoang-Quynh Le","doi":"10.1016/j.jbi.2025.104874","DOIUrl":"10.1016/j.jbi.2025.104874","url":null,"abstract":"<div><h3>Objective:</h3><div>Drug–drug interactions (DDIs) occur when one medication affects the efficacy of another, potentially leading to unforeseen patient outcomes. Existing studies primarily focus on textual data, but overlook a wealth of the drug’s multimodal information. This study aims to enhance DDI extraction by integrating diverse data modalities and evaluating various fusion strategies.</div></div><div><h3>Methods:</h3><div>We introduce a multimodal approach that integrates diverse representations of drug information (scientific text, graphs, formulas, images, and descriptions) to enhance the detection of drug–drug interactions. We explored various fusion techniques to effectively combine these modalities across early, intermediate, and late fusion phases. Additionally, we identify the factors contributing to failed cases, providing insights into the model’s limitations and potential improvements. We have conducted experiments using publicly available DDI datasets, demonstrating significant improvements compared to existing methods.</div></div><div><h3>Results</h3><div>: The proposed model significantly outperformed existing methods in DDI detection. Intermediate fusion strategies, particularly prediction-level concatenation, demonstrated superior accuracy and robustness. Detailed analyses identified factors contributing to failed cases, offering insights for future improvements.</div></div><div><h3>Conclusion:</h3><div>The findings highlight the potential of multimodal fusion to enhance predictive accuracy, providing a foundation for safer drug therapies and better-informed clinical decisions.</div></div>","PeriodicalId":15263,"journal":{"name":"Journal of Biomedical Informatics","volume":"169 ","pages":"Article 104874"},"PeriodicalIF":4.0,"publicationDate":"2025-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144659341","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Identifying homogenous patient subgroups using transformer based hierarchical clustering of heterogeneous Mixed-Modality medical data","authors":"William Baskett ,&nbsp;Benjamin Black ,&nbsp;Adnan I. Qureshi ,&nbsp;Chi-Ren Shyu","doi":"10.1016/j.jbi.2025.104878","DOIUrl":"10.1016/j.jbi.2025.104878","url":null,"abstract":"<div><h3>Objective</h3><div>Patients are highly heterogeneous, with varying needs and responses to treatment. Identifying clinically homogenous patient subgroups is critical to improve personalized care. Patient records are often heterogeneous, may include multiple modalities which conventionally require separate data processing considerations, and are often incomplete, leading to difficulties in identifying meaningful clusters of patients.</div></div><div><h3>Methods</h3><div>We introduce a Med-ROAR, a transformer-based Random Order AutoRegressive (ROAR) embedding model for medical data. Med-ROAR hierarchically clusters data by encoding it into hierarchical discrete embeddings using a modified self-attention operation to facilitate random order mixed modality autoregressive modeling. This allows the model to accept arbitrary mixes of record types without special considerations. We compare our method’s clustering effectiveness to standard agglomerative clustering using 147,469 individuals diagnosed with Autism Spectrum Disorder (ASD). We also evaluate its use on data with mixed modalities and its resilience to missing information using 50,458 clinical records from Intensive Care Unit (ICU) patients which include both tabular and time-series components.</div></div><div><h3>Results</h3><div>We demonstrate that Med-ROAR is more likely to discover more cohesive high-level clusters than distance-based methods like agglomerative clustering. Our exploratory analysis of the autism data identifies clinically meaningful patterns of phenotypes within ASD. We identify homogenous, but atypical, patient subgroups within the ASD population. We also demonstrate Med-ROAR’s effectiveness in clustering patients using mixes of both tabular and time series clinical records from ICU patients. We demonstrate that Med-ROAR can predict patient subgroups even using incomplete, preliminary information collected shortly after admission.</div></div><div><h3>Conclusion</h3><div>Med-ROAR is a flexible hierarchical clustering technique which learns to cluster patients based on learned high-level semantic similarities rather than rule-based metrics. It can accept whatever patient data may be available without modification to the underlying model architecture. The data modalities which Med-ROAR can accept are primarily constrained by computational resources, rather than architectural limitations.</div></div>","PeriodicalId":15263,"journal":{"name":"Journal of Biomedical Informatics","volume":"169 ","pages":"Article 104878"},"PeriodicalIF":4.0,"publicationDate":"2025-07-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144642647","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"COVID-19 prediction with doubly multi-task Gaussian Process","authors":"Sooyon Kim ,&nbsp;Yongtaek Lim ,&nbsp;Sungjun Lim ,&nbsp;Gyeongdeok Seo ,&nbsp;Jihee Kim ,&nbsp;Hojun Park ,&nbsp;Jaehun Jung ,&nbsp;Kyungwoo Song","doi":"10.1016/j.jbi.2025.104872","DOIUrl":"10.1016/j.jbi.2025.104872","url":null,"abstract":"<div><div>This paper addresses a real-world multi-task prediction problem with time-series characteristics by proposing a novel Doubly Multi-Task Gaussian Process (DMTGP) model. Motivated by strong correlations between the number of confirmed cases and deaths, as well as between cases across the different countries, the model incorporates task-wise correlations to predict the number of COVID-19 patients, considering both task-specific (individual) and cross-task (shared) information to enhance overall performance. We constructed a database for three East Asian countries — Japan, South Korea, and Taiwan — and aim to simultaneously predict the number of confirmed cases and deaths in each country. To model the interactions among these countries, we employed a Transformer encoder layer to calculate cross-attention scores. Qualitative analysis of the attention score map demonstrates that our framework effectively captures the dynamic relationships between multiple nations over time. Our experimental results show that the DMTGP model outperforms other baseline models in handling doubly multiple tasks.</div></div>","PeriodicalId":15263,"journal":{"name":"Journal of Biomedical Informatics","volume":"169 ","pages":"Article 104872"},"PeriodicalIF":4.0,"publicationDate":"2025-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144626489","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}