Si Zheng

The etiology of pediatric costochondritis, a common cause of musculoskeletal chest pain, remains poorly understood. This study aimed to identify nutritiona

Microbial symbionts are closely related to the internal and external factors of their host. However, the prevalence of phylosymbiosis (the presence of host phylogenetic signal in microbial community composition) remains controversial, especially in animals collectively referred to as herpetofauna. To expand our understanding of host-microbiota interactions, we analyzed 11,697 symbiotic microbiota samples from of 337 herpetofaunal species, covering skin, oral cavity, gut, cloaca, feces, and other body sites. The composition of the microbial communities gradually changes along the digestive tract, and is host-specific in each region. Overall, herpetofauna’s dominant microbial taxa (Firmicutes, Proteobacteria, Bacteroidota) are more similar to mammals than fish (which are dominated by Proteobacteria, Firmicutes, and Fusobacteriota). However, phylosymbiosis in herpetofauna is weaker than in mammals and tends to occur at higher host taxonomic levels. The strength of the phylosymbiosis signal is influenced by body site, host genetic distance, and analytical method. It indicates that phylosymbiosis exists but is not universal. The intensity and significance of this signal are influenced by host taxonomic scale, the location of the microbial communities, and the the assessment methods. These results advance our knowledge of host-microbe interactions across the Tree of Life.

Transcriptome-wide association studies (TWAS), which combine large-scale GWAS and expression quantitative trait loci (eQTL) datasets to predict gene expression levels and identify trait-related genes, have recently been applied across many traits. TWAS Atlas is a comprehensive public resource that carefully curates TWAS results from a standard set of literature, helping researchers integrate and analyze numerous gene–trait associations. In TWAS Atlas 2.0 (https://ngdc.cncb.ac.cn/twas/), we not only include the latest high-quality TWAS findings but also incorporate GWAS datasets for performing TWAS analysis to expand the range of covered phenotypes, increasing gene–trait associations from 401 226 to 676 198, involving 590 traits and 24 870 genes. The knowledge graph has been significantly enhanced and enriched through information extraction from literature, now featuring new capabilities that greatly improve its usability and analytical power. We have developed multidimensional interactive analysis modules, including functional enrichment, Mendelian randomization, colocalization, and fine-mapping, providing researchers with comprehensive tools to explore causal relationships between genes and traits. These updates significantly enhance the utility of TWAS Atlas 2.0, facilitating a deeper understanding of gene expression and its regulation of complex traits.

Background: Drug repositioning is a pivotal strategy in pharmaceutical research, offering accelerated and cost-effective therapeutic discovery. However, biomedical information relevant to drug repositioning is often complex, dispersed, and underutilized due to limitations in traditional extraction methods, such as reliance on annotated data and poor generalizability. Large language models (LLMs) show promise but face challenges such as hallucinations and interpretability issues. Objective: This study proposed long chain-of-thought for drug repositioning knowledge extraction (LCoDR-KE), a lightweight and domain-specific framework to enhance LLMs’ accuracy and adaptability in extracting structured biomedical knowledge for drug repositioning. Methods: A domain-specific schema defined 11 entities (eg, drug, disease) and 18 relationships (eg, treats, is biomarker of). Following the established schema architecture, we constructed automatic annotation based on 10,000 PubMed abstracts via chain-of-thought prompt engineering. A total of 1000 expert-validated abstracts were curated into a drug repositioning corpus, a high-quality specialized corpus, while the remaining entries were allocated for model training purposes. Then, the proposed LCoDR-KE framework combined supervised fine-tuning of the Qwen2.5-7B-Instruct model with reinforcement learning and dual-reward mechanisms. Performance was evaluated against state-of-the-art models (eg, conditional random fields, Bidirectional Encoder Representations From Transformers, BioBERT, Qwen2.5, DeepSeek-R1, OpenBioLLM-70B, and model variants) using precision, recall, and F₁-score. In addition, the convergence of the training method was assessed by analyzing performance progression across iteration steps. Results: LCoDR-KE achieved an entity F₁ of 81.46% (eg, drug 95.83%, disease 90.52%) and triplet F₁ of 69.04%, outperforming traditional models and rivaling larger LLMs (DeepSeek-R1: entity F₁=84.64%, triplet F₁=69.02%). Ablation studies confirmed the contributions of supervised fine-tuning (8.61% and 20.70% F₁ drop if removed) and reinforcement learning (6.09% and 14.09% F₁ drop if removed). The training process demonstrated stable convergence, validated through iterative performance monitoring. Qualitative analysis of the model’s chain-of-thought outputs showed that LCoDR-KE performed structured and schema-aware reasoning by validating entity types, rejecting incompatible relations, enforcing constraints, and generating compliant JSON. Error analysis revealed 4 main types of mistakes and challenges for further improvement. Conclusions: LCoDR-KE enhances LLMs’ domain-specific adaptability for drug repositioning by offering an open-source drug repositioning corpus and a long chain-of-thought framework based on a lightweight LLM model. This framework supports drug discovery and knowledge reasoning while providing scalable, interpretable solutions applicable to broader biomedical knowledge extraction tasks.

Background: The shortage of pediatric medical resources and overcrowding in children’s hospitals are severe issues in China. Accurately predicting waiting times can help optimize hospital operational efficiency. Objective: This study aims to develop machine learning models to predict waiting times for various laboratory and radiology examinations at a pediatric hospital. Methods: Time stamp data from laboratory and radiology examinations were retrospectively collected from the pediatric hospital information system between November 1, 2024, and March 13, 2025. Two queue-related and 4 time-based features were extracted using queue theory. Linear regression and 8 machine learning models were trained to predict waiting times for each medical task. Hyperparameters were fine-tuned using randomized search and 10-fold cross-validation, and the bootstrap method was used for model evaluation. Mean absolute error, mean square error, root mean square error, and the coefficient of determination (R²) were used as evaluation metrics. Shapley additive explanations values were used to assess feature importance. Results: A total of 230,864 time-stamped records were included after data preprocessing. The median waiting time was 4.817 (IQR 1.867-12.050) minutes for all medical tasks. Waiting times for radiology examinations were generally longer than those for laboratory tests. Tree-based algorithms, such as random forest and classification and regression trees, performed best in predicting laboratory test waiting times, with R² values ranging from mean 0.880(SD 0.003) to mean 0.934 (SD 0.003). However, the machine learning models did not perform well in predicting radiology examination waiting times, with R² ranging from 0.114 (SD 0.005) to 0.719 (SD 0.004). Feature importance analysis revealed that queue-related predictors, especially the number of queuing patients, were the most important in predicting waiting times. Conclusions: Task-specific prediction models are more appropriate for accurately predicting waiting times across various medical tasks. Guided by queue theory principles, we developed machine learning models for the waiting time prediction of each medical task and highlighted the importance of queue-related predictors.

Background: Acute pharmaceutical exposure in children can lead to severe health outcomes and contribute to the inefficient use of medical resources. Objective: This study aimed to investigate the clinical characteristics and outcomes of children with acute pharmaceutical exposure to guide the development of preventive strategies and educational initiatives. Methods: We analyzed real-world data from electronic medical records of children admitted to the emergency department of a pediatric hospital for acute pharmaceutical exposure between January 2019 and December 2023. Clinical data, including laboratory test results, interventions, and outcomes, were collected. We compared different exposure events and conducted logistic regression analysis to identify risk factors for hospitalization. Results: A total of 653 children were included in the study. The most common drugs involved in exposure were vitamins (149/653, 22.8%), nonsteroidal anti-inflammatory drugs (92/653, 14.1%), and psychiatric drugs (74/653, 11.3%). In total, 74.3% (469/631) of patients with complete clinical manifestation data showed no symptoms after exposure, and 68.1% (445/653) of patients did not require specific therapy. Toxicology screening was performed for 11% (72/653) of the children, and 69.4% (50/72) of these tests were positive. Independent risk factors for hospitalization included multisystem involvement (odds ratio [OR] 4.575, 95% CI 1.709-12.251, P=.002), psychiatric drugs (OR 6.280, 95% CI 2.189-18.020, P=.001), and intentional poisoning (OR 12.892, 95% CI 2.222-74.796, P=.004). Conclusions: Children with acute pharmaceutical exposure exhibit diverse clinical characteristics and outcomes, with most requiring no specific treatment. However, immediate toxicology screening and clinical intervention are essential for those exhibiting rapidly developing or multisystem symptoms, as well as those with intentional exposure or exposure to known highly toxic substances. Future pediatric health care policies should emphasize safe storage practices and public education on the prevention of pharmaceutical exposure.

Objective: This study aims to systematically analyze the epidemiological characteristics, clinical interventions and outcomes of children with acute non-pharmaceutical toxic exposures.Methods: This retrospective study included all children with acute non-pharmaceutical toxic exposure admitted to the emergency department of the Capital Institute of Pediatrics between January 1, 2019, and December 31, 2023. Eligible patients were triaged into red, yellow, and green groups according to their severity condition. Clinical information including demographics, exposure details, clinical manifestation, laboratory results, treatments, and outcomes were extracted from electronic medical records. Univariate and multivariate logistic regression analyses were conducted to identify risk factors associated with hospitalization.Results: Overall, a total of 718 patients with acute non-pharmaceutical toxic exposures was included in this study, infants and toddlers accounting for 57.9%. The male-to-female ratio was 1.16:1. The majority exposure events occurred at home (89.3%) and in urban settings (78.4%). Accidental poisoning was the predominant cause, accounting for 94.7%, and the primary route of exposure was oral (93.6%). Mercury was the most common exposed substance, representing 18.8% of cases, particularly among preschool-aged children (31.7%). Patients triaged to red zone had a higher proportion of clinical manifestation and required more aggressive treatments. Although most patients discharged without treatment (78.4%), 19.1% need emergency observation, and 2.5% required hospitalization. Logistic regression analysis showed that corrosive household products exposure (OR = 42.747, 95% CI[5.041–362.520]), skin and mucosal damage (OR = 37.052, 95% CI[5.339–257.153]), pesticides exposure (OR = 33.322, 95% CI[3.863–287.423]), heavy metals exposure (OR = 31.636, 95% CI[1.471–680.210]), neurological manifestation (OR = 22.656, 95% CI[4.766–107.711]), positive toxicology results (OR = 15.105, 95% CI[6.584–34.656]), industrial products exposure (OR = 10.294, 95% CI[1.144–92.641]), and intentional poisoning (OR = 3.060, 95% CI[1.733–5.405]) associated with hospitalization.Conclusion: Pediatric patients exposed to some specific toxicants like industrial products and corrosive household products might associated with a higher risk of severe clinical outcomes. Advocating for enhanced safety regulations or educations and public health initiatives to mitigate the incidence of such exposures is still important for protecting children's health.

Efficiently classifying chronic dizziness disorders, including persistent postural-perceptual dizziness (PPPD), anxiety, and depressive disorders, is crucial, particularly in primary healthcare settings. This study introduces DizzyInsight, an innovative etiological classification model, designed to enhance the accuracy and reliability of large language model (LLM) and machine learning approaches for etiological classification of chronic dizziness. Eight physicians specializing in chronic dizziness diagnosis, affiliated with the Clinical Center for Vertigo and Balance Disturbance at Beijing Tiantan Hospital, Capital Medical University, furnished comprehensive definitions and evaluations of chronic dizziness characteristics. The study included 260 patients, consisting of 105 males and 155 females, with a mean age of 59.52 ± 13 years. These patients were recruited from the same center between July 2021 and October 2023. For comparative analysis, we utilized the general models bidirectional encoder representations from transformers (BERT) and LLM to assess different outcomes. Seven major categories and 33 subcategory evidence have been defined for etiological classification of chronic dizziness. With DizzyInsight, we constructed the feature dataset regarding chronic dizziness. The DizzyInsight based on the identified evidence of LLM method yielded a positive predictive value of 0.69, a sensitivity of 0.86 for persistent postural-perceptual dizziness (PPPD), a positive predictive value of 0.81, and a sensitivity of 0.66 for anxiety and depressive disorders. These findings highlight the potential of DizzyInsight leveraging LLM to improve the efficacy and interpretability of machine learning models in etiological classification of chronic dizziness disorders. Further research and model development are necessary to improve the accuracy of evidence identification and assess the applicability of DizzyInsight in primary care settings, as well as to evaluate its external validity.

Drug resistance in Mycobacterium tuberculosis (Mtb) is a significant challenge in the control and treatment of tuberculosis, making efforts to combat the spread of this global health burden more difficult. To accelerate anti-tuberculosis drug discovery, repurposing clinically approved or investigational drugs for the treatment of tuberculosis by computational methods has become an attractive strategy. In this study, we developed a virtual screening workflow that combines multiple machine learning and deep learning models, and 11 576 compounds extracted from the DrugBank database were screened against Mtb. Our screening method produced satisfactory predictions on three data-splitting settings, with the top predicted bioactive compounds all known antibacterial or anti-TB drugs. To further identify and evaluate drugs with repurposing potential in TB therapy, 15 screened potential compounds were selected for subsequent computational and experimental evaluations, out of which aldoxorubicin and quarfloxin showed potent inhibition of Mtb strain H37Rv, with minimal inhibitory concentrations of 4.16 and 20.67 μM/mL, respectively. More inspiringly, these two compounds also showed antibacterial activity against multidrug-resistant TB isolates and exhibited strong antimicrobial activity against Mtb. Furthermore, molecular docking, molecular dynamics simulation, and the surface plasmon resonance experiments validated the direct binding of the two compounds to Mtb DNA gyrase. In summary, our effective comprehensive virtual screening workflow successfully repurposed two novel drugs (aldoxorubicin and quarfloxin) as promising anti-Mtb candidates. The verification results provide useful information for the further development and clinical verification of anti-TB drugs.