Background: Burkholderia cenocepacia is a threatening nosocomial epidemic pathogen in patients with cystic fibrosis (CF) or a compromised immune system. Its high level of antibiotic resistance is an increasing concern in treatments against its infection. Strain B. cenocepacia J2315 is the most infectious isolate from CF patients. There is a strong demand to reconstruct a genome-scale metabolic network of B. cenocepacia J2315 to systematically analyze its metabolic capabilities and its virulence traits, and to search for potential clinical therapy targets. Results: We reconstructed the genome-scale metabolic network of B. cenocepacia J2315. An iterative reconstruction process led to the establishment of a robust model, i KF1028, which accounts for 1,028 genes, 859 internal reactions, and 834 metabolites. The model i KF1028 captures important metabolic capabilities of B. cenocepacia J2315 with a particular focus on the biosyntheses of key metabolic virulence factors to assist in understanding the mechanism of disease infection and identifying potential drug targets. The model was tested through BIOLOG assays. Based on the model, the genome annotation of B. cenocepacia J2315 was refined and 24 genes were properly re-annotated. Gene and enzyme essentiality were analyzed to provide further insights into the genome function and architecture. A total of 45 essential enzymes were identified as potential therapeutic targets. Conclusions: As the first genome-scale metabolic network of B. cenocepacia J2315, i KF1028 allows a systematic study of the metabolic properties of B. cenocepacia and its key metabolic virulence factors affecting the CF community. The model can be used as a discovery tool to design novel drugs against diseases caused by this notorious pathogen.

Mycobacterial pathogens are a major threat to humans. With the increasing availability of functional genomic data, research on mycobacterial pathogenesis and subsequent control strategies will be greatly accelerated. It has been suggested that genome polymorphisms, namely large sequence polymorphisms, can influence the pathogenicity of different mycobacterial strains. However, there is currently no database dedicated to mycobacterial genome polymorphisms with functional interpretations.

Anti-N-methyl-D-aspartate receptor (anti-NMDAR) encephalitis is an immune-mediated neurological disorder driven by dysregulated neuroimmune interactions, yet the molecular architecture linking tumor-associated immune activation, peripheral immunity, and neuronal dysfunction remains insufficiently understood. In this study, we established an integrative computational framework that combines multi-tissue transcriptomic profiling, weighted gene co-expression network analysis, immune deconvolution, and machine learning-based feature prioritization to systematically characterize the regulatory landscape of the disease. Joint analysis of three independent GEO datasets spanning ovarian teratoma tissue and peripheral blood transcriptomes identified 2001 consistently dysregulated mRNAs, defining a shared tumor–immune–neural transcriptional axis. Across multiple feature selection algorithms, ACVR2B and MX1 were reproducibly prioritized as immune-associated candidate genes and were consistently downregulated in anti-NMDAR encephalitis samples, showing negative correlations with neutrophil infiltration. Reconstruction of an integrated mRNA-miRNA-lncRNA regulatory network further highlighted a putative core axis (ENSG00000262580–hsa-miR-22-3p–ACVR2B), proposed as a hypothesis-generating regulatory module linking non-coding RNA regulation to immune-neuronal signaling. Pathway and immune profiling analyses demonstrated convergence of canonical immune signaling pathways, including JAK-STAT and PI3K-Akt, with neuronal communication modules, accompanied by enhanced innate immune signatures. Although limited by reliance on public datasets and small sample size, these findings delineate a systems-level neuroimmune regulatory program in anti-NMDAR encephalitis and provide a scalable, network-based multi-omics framework for investigating immune-mediated neurological and autoimmune disorders and for guiding future experimental validation.

Background: Differentiating among liver disease entities such as autoimmune liver disease (AILD), drug-induced liver injury (DILI), and chronic hepatitis B (CHB) remains clinically challenging due to overlapping clinical manifestations and nonspecific laboratory findings. Conventional machine learning (ML) approaches rely mainly on structured laboratory data, whereas free-text clinical reports and other heterogeneous electronic medical record data are often underused. Large language models (LLMs) may provide a strategy for encoding heterogeneous clinical information, yet their usefulness for liver disease classification remains insufficiently evaluated. Objective: This study aimed to evaluate the usefulness of LLM-derived embeddings for clinical data mining in liver disease and to determine whether integrating these embeddings with laboratory variables improves classification across broad disease categories and closely related subtypes. Methods: We retrospectively analyzed electronic medical record data from 7543 patients with nonoverlapping liver disease etiologies treated at Beijing Youan Hospital, Capital Medical University, between 2010 and 2025. Three LLMs (Qwen3, Huatuo-o1, and II-Medical) generated semantic embeddings from standardized clinical text, combining free-text examination reports, and structured clinical observations. Performance was assessed in a 3-class etiological task (AILD, DILI, and CHB) and a 4-class task further subclassifying AILD into autoimmune hepatitis and primary biliary cholangitis. We compared embedding-only models, LLM-integrated ML models, and an ML-only baseline using the same structured variable set and preprocessing pipeline, with lightweight natural language processing encoders and zero-shot LLM reasoning as additional comparators. Models were developed using 5-fold cross-validation and evaluated on an internal holdout set using accuracy, macroaveraged precision, recall, and F ₁ -score. Results: In the 3-class task, the LLM-integrated ML models achieved macro F ₁ -scores of 0.835‐0.837, compared with 0.791 for the ML-only baseline, with corresponding accuracies of 0.925‐0.929 versus 0.893. In the 4-class task, the LLM-integrated ML models achieved macro F ₁ -scores of 0.717‐0.734, compared with 0.665 for the ML-only baseline, with corresponding accuracies of 0.920‐0.922 versus 0.874. A temporal split sensitivity analysis using cases from 2010 to 2019 for training and cases from 2020 to 2025 for testing showed that the relative advantage of LLM-integrated ML models over the ML-only baseline was preserved. Direct zero-shot LLM reasoning and lightweight natural language processing encoders performed below the embedding-based integrated models. Conclusions: In this single-center retrospective cohort of patients with clear-cut, nonoverlapping liver disease etiologies, LLM-derived embeddings provided complementary information to structured laboratory variables for multiclass liver disease classification. The integrated framework showed improved internal validation performance compared with the ML-only model, particularly for non-CHB categories and fine-grained subtype discrimination. Because patients with overlapping liver disease etiologies were excluded, the reported performance may overestimate diagnostic accuracy in broader real-world clinical settings where overlapping syndromes are common. Multicenter external validation and prospective evaluation in more heterogeneous patient populations are needed before clinical implementation.

Accurate staging of liver fibrosis in autoimmune hepatitis (AIH) remains challenging due to the invasive nature and sampling limitations of liver biopsy. This study aimed to identify readily available predictors of severe fibrosis and to develop an AIH-specific noninvasive machine-learning model. This two-stage study retrospectively enrolled 208 patients with biopsy-confirmed AIH, with prospective validation in 26 additional patients. Transient elastography (TE) was performed in 110 retrospective and 12 prospective patients. Severe fibrosis was defined as Scheuer stages S3 to S4. Candidate variables underwent univariable and multivariable logistic regression with collinearity control. A random forest (RF) model was trained on the independent predictors and evaluated by the area under the receiver operating characteristic curve (AUROC), calibration, and decision curve analysis. Shapley Additive exPlanations were used for interpretability. Inflammatory activity was graded by Scheuer and prespecified for subgroup analyses (G0–G2 vs G3–G4). A TE-inclusive RF model was also developed in the TE subgroup. The globulin-to-platelet index, international normalized ratio, and blood urea nitrogen were identified as independent predictors of severe fibrosis in AIH. The RF model based on these variables yielded AUROCs of 0.863 (95% confidence interval [CI], 0.802–0.917) in the training set, 0.747 (95% CI, 0.602–0.863) in the test set, and 0.784 (95% CI, 0.556–0.959) in the prospective cohort. Stratified by inflammatory grade, AUROCs were 0.842 (95% CI, 0.757–0.914) in G0–G2 and 0.814 (95% CI, 0.726–0.886) in G3–G4. In contrast, the aspartate aminotransferase-to-platelet ratio index and fibrosis-4 index performed poorly overall and deteriorated further under moderate-to-severe inflammation. In the TE subgroup, the RF model outperformed TE alone (AUROC, 0.786 vs 0.682), and performance improved further when TE was integrated (AUROC, 0.898 [95% CI, 0.841–0.949]). Globulin-to-platelet index, international normalized ratio, and blood urea nitrogen were independent predictors of severe fibrosis in AIH. An RF model constructed from these markers provided a robust, noninvasive tool whose performance was preserved across inflammatory grades and was further enhanced by incorporating TE.