Bacterial infection is a key cause of mortality in patients with acute-on-chronic liver failure (ACLF). In this study, we aimed to identify metabolite biomarkers and develop a novel machine learning model for early identification of bacterial infection in ACLF.

Based on a prospective multicenter cohort from 14 centers, 1,314 patients with acute-on-chronic liver disease were enrolled, including those with ACLF and non-ACLF. Plasma samples at admission were collected for metabolomics profiling. Patients were randomly divided into discovery (n = 921) and validation (n = 393) sets. Machine learning was used to develop diagnostic models. The win ratio method was employed to assess the risk stratification capability of the models.

Bacterial infection occurred in 198 of the 451 ACLF patients and 132 of the 863 non-ACLF patients. Infection altered the plasma metabolome, especially in lipid, amino acid, and xenobiotic metabolic pathways. Models for bacterial infection in ACLF (five metabolites) and non-ACLF (six metabolites) demonstrated superior discrimination in the discovery (AUCs: 0.881 and 0.935, respectively) and validation sets (AUCs: 0.835 and 0.889, respectively) compared with C-reactive protein, white blood cell count, procalcitonin, and the best composite clinical model. Metabolic risk stratification based on the models effectively predicted 90-day outcomes (all-cause death, organ failure, sepsis, new-onset acute decompensation, and systemic inflammatory response syndrome).

Our models based on novel metabolic biomarkers enable identification of patients at high risk of bacterial infection and support risk stratification of 90-day outcomes.

Randomized controlled trials (RCTs) have been conducted to evaluate treatment efficacy for metabolic dysfunction-associated steatotic liver disease (MASLD) and metabolic dysfunction-associated steatohepatitis. This study aimed to compare the effectiveness and safety of 11 promising targets among adults with MASLD.

PubMed, Web of Science, the Cochrane Central Register of Controlled Trials, Scopus, and Embase were searched from inception to November 20, 2024. The primary outcomes were fibrosis improvement ≥1 stage without worsening of steatohepatitis and steatohepatitis resolution without worsening of fibrosis. Additional outcomes included reductions in liver fat content, liver enzymes, metabolic profiles, and selected safety outcomes. The surface under the cumulative ranking curve (SUCRA) was used to rank efficacy.

Of 11,584 articles screened, 44 eligible RCTs (11,410 participants, 33 medications) were included. For fibrosis improvement, d-(R)-pioglitazone (SUCRA: 79.3) and fibroblast growth factor (FGF) 21 analogs (SUCRA: 71.9) ranked higher. For steatohepatitis resolution, glucagon-like peptide-1 (GLP-1)/glucose-dependent insulinotropic polypeptide (GIP) dual receptor agonists (RAs) (SUCRA: 91.7) ranked higher. Co-agonists of GLP-1/GIP/GCG and GLP-1/GCG receptors ranked higher for relative and absolute changes in liver fat content, respectively. For liver enzymes and glucose improvement, the combination of a GLP-1 RA and an acetyl-coenzyme A carboxylase inhibitor ranked higher. GLP-1 RAs, peroxisome proliferator-activated RAs, and FGF21 analogs showed favorable effects on lipid profile improvement.

Incretin-based co-agonists and FGF21 analogs showed favorable profiles across key endpoints, while d-(R)-pioglitazone and GLP-1/GIP dual RAs ranked higher for fibrosis improvement and steatohepatitis resolution, respectively. SUCRA rankings should be interpreted in conjunction with effect sizes, uncertainty, and available safety data.

Hepatic sinusoidal obstruction syndrome (HSOS) is a life-threatening liver vascular disorder with limited treatment options. HSOS results from the activation and injury of liver sinusoidal endothelial cells (LSECs). Berberine (BBR) has been shown to protect endothelial cells in various diseases. However, whether BBR can alleviate liver injury and LSEC disruption in HSOS remains unclear. In this study, we aimed to evaluate the effect of BBR on HSOS.

Two mouse models of HSOS were established using monocrotaline or oxaliplatin. Mice in the treatment groups received a low dose (100 mg/kg) or a high dose (200 mg/kg) of BBR daily. Histology, scanning electron microscopy, immunofluorescence, and flow cytometry were used to evaluate the therapeutic effects of BBR. Cell co-culture, Transwell assays, qRT-PCR, and Western blotting were performed to investigate the molecular pathways involved.

BBR treatment dose-dependently reduced liver injury and disruption of LSECs in murine HSOS models. Moreover, BBR significantly reduced hepatic neutrophil infiltration, thereby attenuating neutrophil-mediated injury to LSECs. Additionally, BBR inhibited the effect of injured LSECs on neutrophil activation. Mechanistically, injured LSECs were identified as one of the major sources of CXCL1 in HSOS, and BBR downregulated CXCL1 expression in injured LSECs by inhibiting MAPK signaling.

In this study, we demonstrate that BBR ameliorates HSOS by inhibiting endothelial-mediated neutrophil recruitment and activation. BBR may be a promising therapeutic option for HSOS treatment.

High-throughput transcriptomic technologies have made differential gene expression analysis a cornerstone of cancer research by generating extensive lists of differentially expressed genes across tumor types and conditions. However, such lists provide limited biological insight without functional and mechanistic interpretation. This review examines the transition from descriptive gene expression profiles to a mechanistic understanding of cancer biology. We discuss integrative approaches that place expression changes within signaling pathways, transcriptional regulatory networks, and protein–protein interaction networks, thereby helping to identify functional modules and candidate upstream regulators. We emphasize the context-dependent nature of gene expression, which is shaped by genetic alterations, epigenetic landscapes, microenvironmental signals, and cellular heterogeneity. We also examine methodological advances, including gene set enrichment analysis, network-based modeling, and integration of genomic, epigenomic, proteomic, metabolomic, and single-cell transcriptomic data. Case studies across cancer types illustrate how mechanistic analyses can reveal context-specific transcriptional programs associated with oncogenic signaling, tumor suppression, metabolic reprogramming, epithelial–mesenchymal transition, and tumor–immune interactions. We highlight potential translational applications, including candidate biomarker discovery, prioritization of druggable targets, rational design of combination therapies, and investigation of therapeutic resistance. Finally, we discuss current challenges and emerging technologies, such as spatial transcriptomics and clustered regularly interspaced short palindromic repeats (CRISPR)-based perturbation screens, that are advancing the field toward dynamic, systems-level models of tumor biology. Integrating computational analyses with experimental validation can help translate transcriptomic data into clinically relevant hypotheses for precision oncology and personalized cancer therapy.

Early predictors of 6-month non-recovery in drug-induced liver injury (DILI) remain limited. Genetic variants are stable host characteristics that may complement baseline clinical variables. We aimed to develop and validate an interpretable, clinical-genetic machine learning model for predicting 6-month non-recovery in patients with DILI.

This retrospective, single-center study included 338 patients with DILI, who were classified as recovered (n = 171) or non-recovered (n = 167) at 6 months. Candidate single-nucleotide polymorphisms and baseline clinical variables were collected during initial hospitalization. Features were selected using complementary screening approaches. Multiple machine learning models were developed and compared. Model discrimination, calibration, clinical utility, the incremental value of genetic predictors, and interpretability using SHapley Additive exPlanations (SHAP) were assessed.

Five predictors were consistently retained for model development: rs72631567, rs28521457, alanine aminotransferase, monocyte percentage, and low-density lipoprotein. Among the candidate algorithms, the light gradient boosting machine model showed the best performance, with area under the receiver operating characteristic curve (AUC) values of 0.92 (95% confidence interval [CI] 0.89–0.95) in the training set and 0.81 (95% CI 0.70–0.91) in the validation set. The model showed acceptable calibration and favorable decision-curve performance. In ablation analysis, the clinical-only model showed limited discrimination (AUC 0.57, 95% CI 0.43–0.71). SHAP analysis identified rs72631567 as the most influential predictor.

An interpretable model that integrates host genetic variants with baseline clinical variables demonstrated good internal performance for early prediction of 6-month non-recovery in DILI. These findings support external validation of genotype-informed risk stratification in patients with DILI.

This study aimed to evaluate the predictive value of cholesterol, high-density lipoprotein, and glucose index (CHG index) alone and combined with the monocyte-to-high-density lipoprotein cholesterol ratio (MHR) or triglyceride-to-high-density lipoprotein cholesterol ratio (TG/HDL-C) for NSTE-ACS.

This cross-sectional diagnostic study included 150 patients with NSTE-ACS and 76 healthy controls. Based on the median Gensini score, patients were divided into high-risk (Gensini score ≥51, n = 75) and low-risk groups (Gensini score <51, n = 75). MHR, TG/HDL-C, and CHG index were compared between patients and controls and between high- and low-risk groups. Their correlations with Gensini scores were assessed. Univariate and Multivariate binary logistic regression analysis was performed to identify factors independently associated with high-risk coronary lesions among patients with NSTE-ACS. The predictive performance of individual indicators (MHR, TG/HDL-C, and CHG index) and their combinations was evaluated using receiver operating characteristic curve analysis.

MHR, TG/HDL-C, and CHG index were significantly higher in the patient group than in the control group (all P < 0.001) and the high-risk group than the low-risk group. Those indicators positively correlated with Gensini scores and were independently associated with high-risk coronary lesions among patients with NSTE-ACS (odds ratio (OR) = 16.051, 95% confidence interval (CI): 13.677-99.650 for MHR; OR = 3.562, 95% CI: 1.868-6.793 for TG/HDL-C; and OR = 2.455, 95% CI: 1.040-5.791 for CHG index). The areas under the curve (AUCs) were 0.803 (95% CI: 0.730-0.876) for MHR, 0.746 (95% CI: 0.666-0.826) for TG/HDL-C, and 0.659 (95% CI: 0.573-0.746) for CHG index. The combination of CHG index and TG/HDL-C achieved an AUC of 0.821 (95% CI: 0.755-0.887), while the combination of CHG index and MHR achieved the higher AUC of 0.872 (95% CI: 0.815-0.929).

MHR, TG/HDL-C, and CHG index are independently associated with high-risk coronary lesions among patients with NSTE-ACS. Combining CHG index with MHR or TG/HDL-C shows numerically higher AUCs for identifying high-risk coronary lesions.

The modern intensive care unit (ICU) inundates clinicians with large volumes of data, leading to cognitive overload and a gap between data availability and actionable insight. While artificial intelligence (AI) promises a solution, its clinical adoption is limited by systemic barriers, including algorithmic bias, a lack of trust, and validation failures. This paper argues that a design philosophy that envisions AI as an autonomous decision-maker, rather than an integrated collaborative tool, has hindered its clinical adoption. We propose an alternative: a collaborative framework designed to augment the intensivist’s expertise by offloading specific cognitive burdens. This framework redefines AI’s purpose as managing data-intensive tasks, illustrated through four collaborative example roles: a synthesizer to create coherent clinical narratives, a sentinel for proactive deterioration surveillance, a simulator to forecast patient responses to interventions, and a stratifier to identify meaningful subphenotypes within complex syndromes. By delegating these computational tasks, this collaborative model frees clinicians to focus on complex synthesis, nuanced judgment, and compassionate communication. Realizing this vision requires a deliberate translational pathway focused on robust data infrastructure, human-centered design, and rigorous validation through prospective clinical trials. Ultimately, the successful integration of AI in critical care depends not on replacing clinicians but on empowering them, creating a more functional ICU in which technology supports the delivery of safer, more precise, and more humane care.