Obesity is a well-established risk factor for major depressive disorder (MDD), yet the risk is not uniform, highlighting the need for precise risk stratification. Our study utilized UK Biobank data from 41,459 obese participants (median follow-up 14.4 years) to develop a metabolomics-based prediction model. By integrating 14 machine learning algorithms, we identified 14 key metabolites that stratify future MDD risk. The optimized LightGBM model achieved AUCs of 0.844, 0.824, and 0.834 for 3-, 5-, and 9-year predictions, significantly outperforming existing clinical models. Mediation Mendelian randomization confirmed causal mediation through specific metabolic pathways. Our findings challenge the notion of a uniform obesity-MDD association, offering a precision medicine approach for early identification of high-risk individuals.

Medical large language models (Med-LLMs) have shown considerable promise across a broad range of clinical tasks, including decision support, medical documentation, patient communication, multimodal analysis, and telemedicine. Their rapid development has generated growing interest in how large language models (LLMs) may support healthcare practice, while also raising important questions about reliability, clinical validity, and safe deployment. This review provides a structured overview of recent progress in Med-LLMs by examining their major application areas, key challenges, and emerging future directions. Current evidence shows that the clinical usefulness of Med-LLMs cannot be judged by model performance alone. Their value in practice depends on whether they are supported by reliable evidence, remain consistent with current medical knowledge, and can be integrated into clinical workflows. Important challenges remain in evaluation, safety, knowledge updating, and real-world deployment. These issues reflect a gap between performance in controlled settings and clinical practice. Future progress will require stronger clinical validation, better alignment with medical practice, and more careful deployment across different settings. The clinical impact of Med-LLMs will depend on whether they can be used as reliable tools in clinical care.

Depression represents a significant global public health challenge, and marital status has been recognized as a potential risk factor. However, previous investigations of this association have primarily focused on Western samples with substantial heterogeneity. Our study aimed to examine the association between marital status and depressive symptoms across countries with diverse cultural backgrounds using a large-scale, two-stage, cross-country analysis. We used nationally representative, de-identified individual-level data from seven countries, including the USA, the UK, Mexico, Ireland, Korea, China and Indonesia (106,556 cross-sectional and 20,865 longitudinal participants), representing approximately 541 million adults. The follow-up duration ranged from 4 to 18 years. Our analysis revealed that unmarried individuals had a higher risk of depressive symptoms than their married counterparts across all countries (pooled odds ratio, 1.86; 95% confidence interval (CI), 1.61–2.14). However, the magnitude of this risk was influenced by country, sex and education level, with greater risk in Western versus Eastern countries (β = 0.36; 95% CI, 0.16–0.56; P < 0.001), among males versus females (β = 0.25; 95% CI, 0.003–0.47; P = 0.047) and among those with higher versus lower educational attainment (β2 = 0.34; 95% CI, 0.11–0.56; P = 0.003). Furthermore, alcohol drinking causally mediated increased later depressive symptom risk among widowed, divorced/separated and single Chinese, Korean and Mexican participants (all P < 0.001). Similarly, smoking was as identified as a causal mediator among single individuals in China and Mexico, and the results remained unchanged in the bootstrap resampling validation and the sensitivity analyses. Our cross-country analysis suggests that unmarried individuals may be at greater risk of depression, and any efforts to mitigate this risk should consider the roles of cultural context, sex, educational attainment and substance use.

Background: Major depressive disorder (MDD) is a leading cause of global disability and poses a substantial public health burden. However, current diagnostic approaches largely rely on subjective assessments and lack the ability to integrate heterogeneous clinical and sociodemographic information. Recent advances in large language models (LLMs) offer new opportunities to support MDD diagnosis through reasoning over complex data, yet their clinical applicability is constrained by challenges related to interpretability, hallucinations, and reliance on synthetic data. Methods: We propose MDD-Thinker, an LLM-based diagnostic system that integrates supervised fine-tuning (SFT) and reinforcement learning (RL) to enhance reasoning and interpretability under the evaluated conditions. Using the UK Biobank dataset, we constructed 40,000 structured reasoning samples and incorporated an additional 10,000 records from publicly available mental health datasets. MDD-Thinker was trained on these heterogeneous textual data and evaluated against conventional machine learning models, deep learning methods, and representative LLM baselines in terms of diagnostic performance and interpretability. Results: MDD-Thinker achieved high performance in MDD diagnosis, with an accuracy of 0.8268 and an F1-score of 0.8081, showing better performance than conventional machine learning models, deep learning approaches, and representative LLM baselines on the evaluated dataset. Beyond predictive accuracy, it consistently produced structured reasoning paths that were clinically coherent, enabling transparent interpretation of diagnostic decisions in the evaluated experiments. The integration of SFT and RL contributed to notable improvements in both diagnostic reliability and reasoning quality. Conclusion: MDD-Thinker demonstrates the potential of reasoning-enhanced LLMs for large-scale MDD diagnosis under the evaluated settings. By jointly optimizing accuracy, interpretability, and efficiency, the proposed system provides a scalable and explainable approach for intelligent psychiatric assessment within the scope of the study, highlighting the potential of reasoning-oriented LLMs in mental health care.

Background: Mendelian randomization (MR) is a powerful epidemiological method that uses genome-wide association study (GWAS) data to infer causal relationships between exposures and outcomes. However, its application is limited by inconsistent data formats, lack of standardized workflows, and the need for programming expertise. To address these challenges, we developed MRanalysis (a user-friendly, web‑based comprehensive MR analysis platform) and GWASkit (a standalone tool for GWAS data preprocessing).

Results: MRanalysis provides a comprehensive, code‑free workflow for MR analysis, including data quality assessment, power estimation, single‑nucleotide polymorphism (SNP)‑to‑gene enrichment analysis, and visualization. It supports univariable, multivariable, and mediation MR analyses through an intuitive interface. GWASkit facilitates rapid preprocessing of GWAS data, such as rs ID conversion and format standardization, with significantly higher accuracy and efficiency than existing tools. Case studies demonstrate the utility and high efficiency of both tools in real‑world scenarios.

Conclusions: MRanalysis and GWASkit lower the barriers to performing MR analysis, making it more accessible, reliable, and efficient. By democratizing MR, these tools can accelerate genetic epidemiological discoveries, inform public health strategies, and guide the development of targeted clinical interventions. MRanalysis is freely available at https://mranalysis.cn, and GWASkit can be accessed at https://github.com/Li-OmicsLab-MPU/GWASkit. Together, they represent a significant advance in understanding the complex relationships among genes, exposures, and health outcomes.