Umeå University's logo

Access to large, diverse biomedical datasets is critical for advancing medical research, yet privacy regulations severely restrict data sharing. We present an end-to-end framework for privacy-preserving health data synthesis that integrates advanced deep generative models (DGMs) with robust preprocessing, formal differential privacy (DP) training for select DGMs, empirical privacy risk evaluation, data-sufficiency analysis, domain-guided quality control, and biobank visualization tools. Released as open-source containerized software, the framework ensures reproducible deployment while preserving statistical fidelity, machine learning (ML) utility, and privacy guarantees. Empirical evaluations across diverse biobank datasets demonstrate that TabSyn—a transformer-based diffusion model–combined with our correlation—and distribution-aware CorrDst loss function achieves superior performance balancing fidelity, privacy, and computational efficiency. The tailored preprocessing pipeline effectively handles high missingness rates, substantially improving distributional accuracy and clinical plausibility. Across 26 biobank datasets spanning three regulatory levels, the framework shows that TabSyn with correlation- and distribution-aware loss function consistently achieves superior performance in terms of fidelity, privacy, and computational efficiency.

Purpose: Antidepressant use, which targets intracerebral serotonin metabolism, is common among glioblastoma patients. However, its interaction with tumor metabolism and survival remains unclear. We investigated the relationships between antidepressant use, intratumoral serotonin pathway metabolites, quality-of-life, and survival.

Methods: We analysed two complementary cohorts: a population-based cohort (n = 801) and a hospital-based biobank cohort (n = 153). In the population-based cohort, information about antidepressant use and survival were obtained from national registers. In the hospital-based cohort, intratumoral serotonin pathway metabolites were measured, and data on preoperative antidepressant use and quality-of-life were recorded. Linear and Cox regression models were used to investigate the association between antidepressant use, metabolites, quality-of-life and survival.

Results: In the population-based cohort, antidepressant use was associated with poorer survival in adjusted analyses. However, use of fluoxetine or sertraline was associated with better survival compared with other selective serotonin reuptake inhibitors (HR = 0.62, 95% CI = 0.44-0.88). In the hospital-based cohort, preoperative antidepressant use was associated with lower intratumoral serotonin levels and its downstream metabolite, 5-HIAA. Higher serotonin levels were associated with better preoperative quality-of-life, especially general health. Serotonin pathway metabolites were not clearly associated with survival.

Conclusions: These findings suggest that higher tumor tissue serotonin abundance was associated with better patient-reported well-being but not with survival in gliobastoma. Survival differed across SSRI exposure groups, although causal interpretation is limited by the observational design. These findings do not provide strong evidence against the continued clinical use of sertraline or fluoxetine in glioblastoma patients when indicated.

Background: Gliomas are the most common malignant primary tumor of the central nervous system and show a high mortality, particularly at higher grades. Cancer predisposition syndromes and common low-penetrance single nucleotide polymorphisms have been shown to contribute to glioma risk, but the contribution of rare germline variants remains incompletely understood. Here, we investigated rare germline variants in glioma patients.

Methods: We performed whole-genome sequencing on 113 glioma patients from Northern Sweden, analyzing rare germline variants across 651 genes. Variants were compared to population controls (ACpop, gnomAD) and validated in TCGA glioma data, a UK Biobank glioma nested case–control study, and a separate cohort of 105 Swedish glioblastomas.

Results: 17.6% of glioma cases carried a Pathogenic or Likely Pathogenic (P/LP) variant within 1 of the 651 genes, and the number of alleles carrying a P/LP was significantly more than in the reference data (P = 3.2 × 10^-3). Many of the observed candidate genes also harbored P/LP variants in our Swedish validation cohort. Overall, gene-based comparison of rare coding variants indicated an enrichment in several genes, including TP53, CREBBP, and DNMT3A.

Conclusions: Rare P/LP germline variants were more frequent among glioma patients than in the reference population within our predefined gene set. These results suggest a contribution of rare germline variants to glioma risk, particularly in genes involved in DNA repair. While several genes are indicated as enriched with rare variants, only TP53 validates across all 3 patient sets.

While longer genetically predicted leukocyte telomere length (LTL) has been linked to increased glioma risk in adults, this association has not been investigated in pediatric populations. In this study, we applied Mendelian randomization (MR) and polygenic risk score (PRS) analyses to investigate the relationship between LTL and pediatric glioma risk, using 4,069 cases and 8,778 controls from the largest available childhood glioma meta-GWAS. Results suggested longer genetically predicted LTL was significantly associated with increased childhood glioma risk, with OR per 1 standard deviation increase in LTL of 2.12 (95% CI 1.32–3.39, P =.002) in the multi-ancestry group and 2.16 (95% CI 1.16–4.03; P =.015) in the European group. Key SNPs contributing to risk of childhood glioma included rs59294613 (POT1), rs8105767 (ZNF208), and rs7705526 (TERT), which differed from the top variants previously identified in adult glioma using the same genetic instruments. Age-stratified MR revealed a stronger association in children diagnosed after the age of 6 (OR = 1.89; P <.001) vs. ≤ 6 years (OR = 1.04; P =.732; Phet =.006). PRS analysis further supported this age-stratified findings by demonstrating a positive association between LTL PRS and age at glioma diagnosis, particularly within age 0–10. This trend suggests a progressively greater influence of LTL on glioma risk in older children compared to younger children. In conclusion, this study provides the first genetic evidence linking longer genetically predicted LTL to increased pediatric glioma risk. While overall consistent with adult findings, distinct single-SNP associations, and age-dependent effects highlight unique biological mechanisms in children with glioma, warranting further direct investigation into telomere dynamics in early-life gliomagenesis.

Deep learning (DL) models require extensive data to achieve strong performance and generalization. Deep generative models (DGMs) offer a solution by synthesizing data. Yet current approaches for tabular data often fail to preserve feature correlations and distributions during training, struggle with multi-metric hyperparameter selection, and lack comprehensive evaluation protocols. We address this gap with a unified framework that integrates training, hyperparameter tuning, and evaluation. First, we introduce a novel correlation- and distribution-aware loss function that regularizes DGMs, enhancing their ability to generate synthetic tabular data that faithfully represents the underlying data distributions. Theoretical analysis establishes stability and consistency guarantees. To enable principled hyper-parameter search via Bayesian optimization (BO), we also propose a new multi-objective aggregation strategy based on iterative objective refinement Bayesian optimization (IORBO), along with a comprehensive statistical testing framework. We validate the proposed approach using a benchmarking framework with twenty real-world datasets and ten established tabular DGM baselines. The correlation-aware loss function significantly improves the synthetic data fidelity and downstream machine learning (ML) performance, while IORBO consistently outperforms standard Bayesian optimization (SBO) in hyper-parameter selection. The unified framework advances tabular generative modeling beyond isolated method improvements. Code is available at: https://github.com/vuhoangminh/TabGen-Framework.

Serotonin and serotonin metabolism has for decades been understood as playing a critical role in mood disorders and has more recently also been implicated in brain tumour biology. However, in part due to the lack of direct investigation of genetic and epigenetic variation affecting serotonin pathways within human brain tissue this understanding has recently been challenged. We analysed genetic and epigenetic variation in the Monoamine oxidase A (MAOA) and serotonin transporter (5HTT) genes using 232 biobanked glioma tissue samples from 216 adult patients. We further examined the association between use of antidepressants (targeting serotonergic pathways), serotonin levels and methylation. In male patients, genetic variation in the MAOA gene was significantly associated with tissue serotonin levels. Further analysis identified five single nucleotide variants (SNVs) that may contribute to this association. In contrast, 5HTT variants were not statistically associated with serotonin pathway metabolites, nor were MAOA variants in females. Increased methylation at several 5HTT CpG sites was positively correlated with serotonin levels and negatively correlated with 5-HIAA levels. In males, one CpG site in the MAOA gene was negatively associated with the 5-HIAA/serotonin ratio, suggesting reduced enzymatic degradation of serotonin due to lower MAOA activity. Patients using antidepressants had lower tissue serotonin levels. In males, genetic variation in the MAOA gene was significantly associated with tissue serotonin levels, although this association was not mediated by methylation. Our result supports the notion that the MAOA and 5HTT genes are related to serotonin metabolism and that such metabolism is related to antidepressant use.

Key features: 

• The Northern Sweden Health and Disease Study (NSHDS) was initiated in the mid-1980s. The NSHDS is a population-based prospective longitudinal cohort comprising >140 000 participants in the two northernmost regions in Sweden, Norrbotten and Västerbotten, with >240 000 blood samples and 1.5 million person-years of follow-up.
• The NSHDS includes three sub-cohorts: the Västerbotten Intervention Programme (VIP), the expanded Northern Sweden Monitoring of Trends and Determinants of Cardiovascular Disease (MONICA) Study, and the Mammography Screening Project (MSP). The VIP is both a community-based cardiometabolic intervention programme encouraging healthy lifestyle (targeting individuals 40, 50, and 60 years of age), and a corresponding research cohort. The MONICA is an observational study focusing on cardiovascular disease and its associated risk factors, recruiting individuals aged 25–74 years. The MSP recruited women attending mammography during 1995–2006. The NSHDS median participation age is 50 years (53% women).
• Most participants contribute data on health, lifestyle, anthropometric measures, blood pressure, blood lipids, and glucose tolerance, along with research blood samples that are fractionated, frozen within an hour of collection, and stored at –80°C. Linkage to registries, clinical cohorts, and biological tissue archives facilitates studies of well-characterized participants (often combined with intervention studies).
• Collaborations are encouraged. Additional information can be found at: info.brs@umu.se; https://www.umu.se/en/biobank

This review explores the pivotal role of preanalytical variables in bringing liquid biopsy approaches into the clinic for brain tumors. Preanalytical variables encompass a range of critical issues, from blood sample collection and handling to the impact of tumor heterogeneity and patient-specific factors. These variables introduce challenges such as false positives, false negatives, and variability in the analysis of tumor signals, which can hinder the diagnostic and prognostic utility of liquid biopsies. Understanding the nuances of preanalytical variables is essential for the successful implementation of liquid biopsy in clinical settings. This paper delves into strategies aimed at mitigating the influence of preanalytical variables by emphasizing the importance of standardized sample collection protocols, optimized sample processing and storage, quality control measures, and the integration of multiple liquid biopsy modalities.

Purpose: Although sociodemographic factors such as socioeconomic status (SES), travel time to health care, cohabitation status, and region of residence are observed to influence incidence and survival for several types of cancers, it is unclear whether similar effects have been observed in patients with glioma. This study investigates whether these factors affect survival for glioma patients.

Methods: In this retrospective study, the Swedish National Quality Registry for Brain Tumors was used to identify 1,276 patients with glioma WHO grade I–IV for whom data were deposited between 2009 and 2013. The RISK North database, which links data from the National Cancer Quality Register with citizen demographic data from the Longitudinal Integration Database for Health Insurance and Labor Market Studies (LISA), the Total Population Registry (TPR), and the Geography Database (GD), was utilized to assess survival in patients with glioma in relation to education level, cohabitation status, travel time to regional hospitals, and region of residence.

Results: In the multivariable analysis, longer survival was observed among WHO grade III-IV glioma patients with higher education level (middle school (ref) HR: 1, high school HR: 0.81 CI [0.67–0.98], p = 0.033; university/college HR: 0.81 CI [0.66–1.00], p = 0.048). Survival was not associated with travel time, cohabitation status, or region of residence in the multivariable survival analysis.

Conclusion: Low education level was associated with reduced survival for patients with glioma WHO grade III and IV in multivariable survival analyses, but no differences in survival were found in relation to travel time, cohabitation status, or region of residence.