Computational and structural biotechnology journal最新文献

{"title":"Comparative analysis of machine learning techniques in metabolomic-based preterm birth prediction.","authors":"Ying-Chieh Han, Jane Shearer, Chunlong Mu, Donna M Slater, Suzanne C Tough, Gavin E Duggan","doi":"10.1016/j.csbj.2025.07.010","DOIUrl":"10.1016/j.csbj.2025.07.010","url":null,"abstract":"<p><strong>Background: </strong>Machine learning (ML), with advancements in algorithms and computations, is seeing an increased presence in life science research. This study investigated several ML models' efficacy in predicting preterm birth using untargeted metabolomics from serum collected during the third trimester of gestation.</p><p><strong>Methods: </strong>Samples from 48 preterm and 102 term delivery mothers from the All Our Families Cohort (Calgary, AB) were examined. Four ML algorithms: Partial Least Squares Discriminant Analysis (PLS-DA), linear logistic regression, artificial neural networks (ANN), Extreme Gradient Boosting (XGBoost) - with and without bootstrap resampling were used to examine the small-scale clinical dataset for both model performance and metabolite interpretation.</p><p><strong>Results: </strong>Model performance was evaluated based on confusion matrices, area under the receiver operating characteristic (AUROC) curve analysis, and feature importance rankings. Linear models such as PLS-DA and logistic regression demonstrated moderate classification performance (AUROC ≈ 0.60), whereas non-linear approaches, including ANN and XGBoost, exhibited marginal improvements. Among all models, XGBoost combined with bootstrap resampling achieved the highest performance, yielding an AUROC of 0.85 (95 % CI: 0.57-0.99, p < 0.001), indicating a significant improvement in classification accuracy. Metabolite importance, derived from Shapley Additive Explanations (SHAP), consistently identified acylcarnitines and amino acid derivatives as principal discriminative features. Pathway analysis revealed disruptions to tyrosine metabolism as well as phenylalanine, tyrosine and tryptophan biosynthesis to be associated with preterm delivery.</p><p><strong>Conclusions: </strong>Our results highlight the complexity of metabolomics-based modelling for preterm birth and support an iterative, model-driven approach for optimizing predictive accuracy in small-scale clinical datasets.</p>","PeriodicalId":10715,"journal":{"name":"Computational and structural biotechnology journal","volume":"27 ","pages":"3240-3250"},"PeriodicalIF":4.1,"publicationDate":"2025-07-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12312043/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144759371","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A comprehensive review on computational metabolomics: Advancing multiscale analysis through <i>in-silico</i> approaches.","authors":"Mohamed S Nafie, Abdelghafar M Abu-Elsaoud, Mohamed K Diab","doi":"10.1016/j.csbj.2025.07.016","DOIUrl":"10.1016/j.csbj.2025.07.016","url":null,"abstract":"<p><p>Computational metabolomics will be established in drug discovery and research on complex biological networks. This field of research enhances the detection of metabolic biomarkers and the prediction of molecular interactions by combining multiscale analysis with <i>in silico</i> and molecular docking methods. These include nuclear magnetic resonance, mass spectrometry, and innovative bioinformatics, which enable the accurate generation and characterization of metabolomes. Molecular docking is a crucial tool for simulating the interaction between ligands and receptors, thereby facilitating the identification of potential therapeutics. It also discusses the potential of metabolomics to inform drug modes of action, from pharmacokinetics to forecasting toxicity, thereby streamlining drug development pipelines. We highlight applications in anticancer, antimicrobial, and antiviral drug discovery and explain how these computational models can accelerate target validation and enhance the accuracy of therapeutic strategies. In addition, this review addresses the current challenges and future directions for computational techniques in conjunction with experimental data to advance personalized medicine. In conclusion, this review aims to highlight the prospective approaches of computational metabolomics and molecular docking that identify evolutionary adaptive metabolisms of multiscale biological systems through their synergistic utilization to overcome the key hurdles involved in both drug discovery and metabolomic research.</p>","PeriodicalId":10715,"journal":{"name":"Computational and structural biotechnology journal","volume":"27 ","pages":"3191-3215"},"PeriodicalIF":4.1,"publicationDate":"2025-07-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12305607/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144741430","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Review of protein structure-based analyses that illuminate plant stress mechanisms.","authors":"Fatima Shahid, Neeladri Sen, Hawa Najibah Rasni, Nurulhikma Md Isa, Nyuk Ling Ma, Christine Orengo, Su Datt Lam","doi":"10.1016/j.csbj.2025.07.021","DOIUrl":"10.1016/j.csbj.2025.07.021","url":null,"abstract":"<p><p>Plants face formidable challenges due to environmental stresses, including pathogens, salt, drought, cold, heat, heavy metal exposure, and flooding, all of which affect growth and agricultural productivity. To combat these stresses, plants have evolved various adaptive mechanisms, including the expression of stress-response proteins. Exploring the three-dimensional structures of plant proteins can be valuable for discovering and characterising stress tolerance mechanisms at the molecular level. Until recently, large-scale analyses were not feasible due to the limited number of experimentally determined plant protein structures. However, the recently developed AlphaFold, RoseTTA-Fold, and ESM-fold protein structure prediction methods, along with their associated portals, now provide hundreds of millions of high-quality predicted 3D models, covering a wide range of plant proteins. This review highlights insights from recent structural investigations into plant stress response using experimental or predicted protein structures. We include analyses of diverse paralogs and isoforms and insights from molecular docking and molecular dynamics simulations. We consider the value of using experimental and predicted structural data in understanding the mechanisms of common stress-modulating plant proteins. Studying the structures of these proteins together with their inferred functions can aid improvements in crop productivity, help foster sustainable agriculture, and contribute to global food security efforts.</p>","PeriodicalId":10715,"journal":{"name":"Computational and structural biotechnology journal","volume":"27 ","pages":"3155-3166"},"PeriodicalIF":4.1,"publicationDate":"2025-07-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12302779/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144728499","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"LC-QTOF-MS^E with MS¹-based precursor ion quantification and SiMD-assisted identification enhances human urine metabolite analysis.","authors":"Alongkorn Kurilung, Suphitcha Limjiasahapong, Kwanjeera Wanichthanarak, Weerawan Manokasemsan, Khwanta Kaewnarin, Kassaporn Duangkumpha, Siriphan Manocheewa, Rossarin Tansawat, Roongruedee Chaiteerakij, Intawat Nookaew, Yongyut Sirivatanauksorn, Sakda Khoomrung","doi":"10.1016/j.csbj.2025.07.009","DOIUrl":"10.1016/j.csbj.2025.07.009","url":null,"abstract":"<p><p>This study presents the development and validation of a liquid chromatography-quadrupole-time-of-flight mass spectrometry method with data-independent acquisition (LC-QTOF-MS^E) for targeted quantification, post-targeted screening, and untargeted metabolite profiling. Using MS¹-based precursor ion quantification, the method demonstrated excellent analytical performance with linearity (<i>R</i>² > 0.99), accuracy (84 %-131 %), and precision (1 %-17 % relative standard deviation (RSD)). Although LC-QTOF‑MS^E sensitivity is at least nine-fold lower than LC-triple quadrupole MS with multiple reaction monitoring, it remains adequate for quantifying urinary metabolites, particularly those that fragment poorly or yield low‑intensity product ions. For post‑targeted screening and untargeted profiling, an in‑house reference library (the Siriraj Metabolomics Data Warehouse, SiMD), comprising 174 curated metabolite standards, was integrated into the workflow to enhance metabolite identification confidence. The official website for SiMD can be accessed at https://si-simd.com/. To demonstrate the method's utility, 11 amino and organic acids were quantified in urine samples from 100 healthy individuals. Four compounds-L-methionine, L-histidine, L-tryptophan, and <i>trans</i>-ferulic acid-were significantly higher levels in females (<i>P</i> < 0.05), likely reflecting sex-specific physiological or dietary intake differences. Post‑targeted screening identified 29 additional metabolites and assigned them to level 1 (<i>m</i>/<i>z</i>, RT, isotope pattern, and MS/MS spectra matched to reference standards) based on the Metabolomics Standards Initiative guidelines. Untargeted retrospective profiling revealed level 1 seven metabolites, including ribitol, creatine, glucuronic acid, <i>trans</i>-ferulic acid, succinic acid, dimethylglycine, and 3-hydroxyphenylacetic acid related to sex variation (VIP > 1.5). In summary, the LC-QTOF-MS^E method coupled with SiMD provides a robust and comprehensive workflow for metabolomics analysis. It enables reliable target quantification and enhances confidence in metabolite identification while also reducing sample and instrumental demands. These features make it particularly well-suited for clinical metabolomics studies.</p>","PeriodicalId":10715,"journal":{"name":"Computational and structural biotechnology journal","volume":"27 ","pages":"3079-3089"},"PeriodicalIF":4.4,"publicationDate":"2025-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12284563/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144697832","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Structural insights into Beclin 1 interactions with it's regulators for autophagy modulation.","authors":"Debapriyo Sarmadhikari, Shailendra Asthana","doi":"10.1016/j.csbj.2025.06.044","DOIUrl":"10.1016/j.csbj.2025.06.044","url":null,"abstract":"<p><p>The molecular recognition process between proteins is the foundation of complex biological functions, driven by residue-level interactions between regulatory and functional domains. Therefore, change in network is the root cause of normal physiology to pathophysiology. Since the network can only be traced through structural data, such insights are essential. However, identifying the critical structural and conformational determinants facilitating signalling cascades remains a major challenge for protein-protein interactions (PPIs) based therapeutic interventions. This challenge is further compounded by the absence of structural data, which makes deciphering the intricate web of PPIs even more difficult. Structural insights are paramount, as PPIs are inherently flexible, exploring a dynamic conformational space characterized by low-energy states interconnected by high-energy transition paths. Autophagy is a cellular process heavily reliant on PPIs, and researchers from academia and industry are targeting them for therapeutic intervention due to their beneficial role in the modulation of multiple diseases, including cancer, neurodegenerative and metabolic diseases. In autophagy pathway, Beclin 1 is a pivotal protein in the signalling cascade. However, targeting Beclin 1 for therapeutic purposes and understanding its role in the signalling cascades remain challenging, primarily due to the lack of structural insights into the mechanisms governing its interactions with its regulatory partners. To overcome these challenges, we integrate AlphaFold predicted models with experimentally resolved PDB structures to construct a comprehensive, domain wise and residue level map of Beclin 1 interactome capturing both structured and unstructured regions, identifying critical interaction interfaces, and uncovering pivotal determinants for Beclin 1 specific therapeutic interventions.</p>","PeriodicalId":10715,"journal":{"name":"Computational and structural biotechnology journal","volume":"27 ","pages":"3005-3035"},"PeriodicalIF":4.4,"publicationDate":"2025-07-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12275485/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144674058","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A single HIIT session does not alter blood sphingolipid levels in healthy young adults: The SphingoHIIT randomized controlled trial.","authors":"Thomas Angst, Nadia Weber, Seraina Fischer, Joëlle Lehmann, Denis Infanger, Tony Teav, Fabian Schwendinger, Lukas Streese, Timo Hinrichs, Ilaria Croci, Christian Schmied, Hector Gallart-Ayala, Christoph Höchsmann, Karsten Koehler, Henner Hanssen, Arno Schmidt-Trucksäss, Julijana Ivanisevic, Justin Carrard","doi":"10.1016/j.csbj.2025.07.004","DOIUrl":"10.1016/j.csbj.2025.07.004","url":null,"abstract":"<p><strong>Introduction: </strong>Sphingolipids and ceramides have been identified as critical drivers of cardiometabolic diseases. Ceramide-based scores were developed, predicting cardiometabolic risk independently of and beyond low-density lipoprotein cholesterol. To date, it remains largely unknown whether exercise can modulate sphingolipid levels.</p><p><strong>Methods: </strong>The SphingoHIIT study was the first parallel randomized controlled trial to investigate the impact of a single session of high-intensity interval training (HIIT; 4 ×4 min at 85-95 % of maximal heart rate) on blood sphingolipid levels. Thirty-six healthy young individuals (aged 20-29 years; 50 % female) were randomly assigned to a HIIT (n = 18) or control group (physical rest, n = 18). Sphingolipid levels were measured from dried blood spots collected over three days before and at five time points after the intervention (2, 15, 30, 60 min, and 24 h). Study conditions were tightly controlled: females were tested during the early follicular phase of their menstrual cycle, and standardized meals were provided for four consecutive days before blood sampling.</p><p><strong>Results: </strong>Forty-seven sphingolipid species were acquired, including 25 ceramides, eight glycosphingolipids, eight sphingomyelins, and six sphingoid bases. After adjusting for sex, body fat mass, cardiorespiratory fitness, and daily physical activity, linear mixed models showed no significant differences in sphingolipid levels between the HIIT and control groups at any post-intervention time point.</p><p><strong>Conclusion: </strong>The present findings suggest that circulating sphingolipids are resilient to an acute bout of intensive exercise, an interesting feature for potential biomarkers of cardiometabolic risk. Future studies should investigate whether regular exercise influences sphingolipid levels and improves cardiometabolic health in different clinical populations.</p>","PeriodicalId":10715,"journal":{"name":"Computational and structural biotechnology journal","volume":"27 ","pages":"2976-2989"},"PeriodicalIF":4.4,"publicationDate":"2025-07-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12274307/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144674049","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"konnect2prot 2.0: Integrating advanced analytical tools for deeper understanding of protein properties in a functional protein-protein interaction network.","authors":"Shivam Kumar, Abhinav Agarwal, Dipanka Tanu Sarmah, Samrat Chatterjee","doi":"10.1016/j.csbj.2025.07.006","DOIUrl":"10.1016/j.csbj.2025.07.006","url":null,"abstract":"<p><p>Proteins work in coordination to catalyze and regulate all biological activities and drive cellular functions. The collective behavior of proteins is studied through protein-protein interaction (PPI) networks, which provide a system-level understanding of their regulatory behavior. konnect2prot (k2p) is a web application that generates a context-specific directional PPI network from a list of proteins as input. It identifies influential spreaders in the generated network and detects their biological and topological importance. Here, we report the features added to k2p since its first release. Building on the foundation of its predecessor, k2p 2.0 now integrates differential gene expression analysis, thereby bridging the gap between gene-level regulation and protein-level activity, providing a holistic view of how transcriptional changes fuel cellular behavior. konnect2prot 2.0 is freely accessible at: https://konnect2prot_v2.thsti.in.</p>","PeriodicalId":10715,"journal":{"name":"Computational and structural biotechnology journal","volume":"27 ","pages":"3036-3044"},"PeriodicalIF":4.4,"publicationDate":"2025-07-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12275489/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144674054","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Inhibiting MARCH5/Mfn2 signaling as an alternative strategy to protect cardiomyocytes from hypoxia-induced mitochondrial dysfunction.","authors":"Faten Habrat Zoabi, Mulate Zerihun, Roy Lizarovich, Chiara Dalla Torre, Liron Davis, Offir Ertracht, Michal Barsheshet, Shaul Atar, Deborah E Shalev, Marta De Zotti, Hanoch Senderowitz, Nir Qvit","doi":"10.1016/j.csbj.2025.07.001","DOIUrl":"10.1016/j.csbj.2025.07.001","url":null,"abstract":"<p><p>The mitochondrial E3 ubiquitin ligase membrane-associated RING-CH-type finger 5 (MARCH5) and the GTPase Mitofusin 2 (Mfn2) both play crucial roles in regulating mitochondrial dynamics, which are essential for cellular homeostasis. Dysregulation of the MARCH5/Mfn2 signaling has been implicated in mitochondrial dysfunction, a key factor in cardiovascular diseases (CVDs). To investigate the therapeutic potential of targeting this interaction, we developed a novel peptide, CVP-220, designed to specifically disrupt the MARCH5/Mfn2 protein interaction. Using a hypoxia-reoxygenation (H/R) injury model in rat cardiomyocyte cell lines, CVP-220 demonstrated significant cardioprotective effects. Treatment with CVP-220 enhanced cell viability by 30 % compared to untreated controls and reduced reactive oxygen species (ROS) production by 45 %, suggesting improved mitochondrial function. Notably, CVP-220 selectively modulated MARCH5-mediated ubiquitination of Mfn2 without affecting other MARCH5 interactions, thereby preserving mitochondrial fusion and preventing fragmentation under stress conditions. A plausible binding mode of CVP-220 on Mfn2 was suggested through a combination of molecular docking and molecular dynamics simulations and was experimentally validated by mutational analysis. These findings highlight CVP-220 as a promising tool for modulating mitochondrial dynamics and mitigating mitochondrial damage in cardiac cells, with potential implications for therapeutic strategies targeting mitochondrial dysfunction in CVDs. Further investigation into the role of MARCH5/Mfn2 signaling in cardiac pathology could pave the way for novel peptide-based treatments.</p>","PeriodicalId":10715,"journal":{"name":"Computational and structural biotechnology journal","volume":"27 ","pages":"3045-3065"},"PeriodicalIF":4.4,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12275895/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144674053","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Leveraging multiple labeled datasets for the automated annotation of single-cell RNA and ATAC data.","authors":"Joseba Sancho-Zamora, Akash Kanhirodan, Xabier Garrote, Juan Manuel Silva Rojas, Olivier Gevaert, Mikel Hernaez, Guillermo Serrano, Idoia Ochoa","doi":"10.1016/j.csbj.2025.06.043","DOIUrl":"10.1016/j.csbj.2025.06.043","url":null,"abstract":"<p><p>The creation of single-cell atlases is essential for understanding cellular diversity and heterogeneity. However, assembling these atlases is challenging due to batch effects and the need for accurate and consistent cell annotation. Current methods for single-cell RNA and ATAC sequencing (scRNA-Seq and scATAC-Seq), while effective for integration, are not optimized for cell annotation. Additionally, many annotation tools rely on external databases or reference scRNA-Seq datasets, which may limit their adaptability to specific study needs, especially for rare cell-types or scATAC-Seq data. Here, we introduce JIND-Multi, a new framework designed to transfer cell-type labels across multiple annotated datasets. Notably, JIND-Multi can be applied to both scRNA-Seq and scATAC-Seq data, requiring in each case annotated data of the same type, contrary to most methods for scATAC-Seq data that require (paired) annotated scRNA-Seq data. In both cases, JIND-Multi significantly reduces the proportion of unclassified cells while maintaining the accuracy and performance of the original JIND model, and compares favorable to state-of-the-art methods. These results prove its versatility and effectiveness across different single-cell sequencing technologies. JIND-Multi represents an improvement in cell annotation, reducing unassigned cells and offering a reliable solution for both scRNA-Seq and scATAC-Seq data. Its ability to handle multiple labeled datasets enhances the precision of annotations, making it a valuable tool for the single-cell research community. JIND-Multi is publicly available at: https://github.com/ML4BM-Lab/JIND-Multi.git.</p>","PeriodicalId":10715,"journal":{"name":"Computational and structural biotechnology journal","volume":"27 ","pages":"2863-2870"},"PeriodicalIF":4.4,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12270792/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144674056","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing glucose level prediction of ICU patients through hierarchical modeling of irregular time-series.","authors":"Hadi Mehdizavareh, Arijit Khan, Simon Lebech Cichosz","doi":"10.1016/j.csbj.2025.06.039","DOIUrl":"10.1016/j.csbj.2025.06.039","url":null,"abstract":"<p><p>Accurately predicting blood glucose (BG) levels of ICU patients is critical, as both hypoglycemia (BG < 70 mg/dL) and hyperglycemia (BG > 180 mg/dL) are associated with increased morbidity and mortality. This study presents a proof-of-concept machine learning framework, the Multi-source Irregular Time-Series Transformer (MITST), designed to predict BG levels in ICU patients. In contrast to existing methods that rely heavily on manual feature engineering or utilize limited Electronic Health Record (EHR) data sources, MITST integrates diverse clinical data-including laboratory results, medications, and vital signs-without predefined aggregation. The model leverages a hierarchical Transformer architecture, designed to capture interactions among features within individual timestamps, temporal dependencies across different timestamps, and semantic relationships across multiple data sources. Evaluated using the extensive eICU database (200,859 ICU stays across 208 hospitals), MITST achieves a statistically significant ( <math><mi>p</mi> <mo><</mo> <mn>0.001</mn></math> ) average improvement of 1.7 percentage points (pp) in AUROC and 1.8 pp in AUPRC over a state-of-the-art random forest baseline. Crucially, for hypoglycemia-a rare but life-threatening condition-MITST increases sensitivity by 7.2 pp, potentially enabling hundreds of earlier interventions across ICU populations. The flexible architecture of MITST allows seamless integration of new data sources without retraining the entire model, enhancing its adaptability for clinical decision support. While this study focuses on predicting BG levels, we also demonstrate MITST's ability to generalize to a distinct clinical task (in-hospital mortality prediction), highlighting its potential for broader applicability in ICU settings. MITST thus offers a robust and extensible solution for analyzing complex, multi-source, irregular time-series data.</p>","PeriodicalId":10715,"journal":{"name":"Computational and structural biotechnology journal","volume":"27 ","pages":"2898-2914"},"PeriodicalIF":4.4,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12270796/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144674051","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}