Stratified neural networks in a time-to-event setting.

Deep neural networks are frequently employed to predict survival conditional on omics-type biomarkers, e.g., by employing the partial likelihood of Cox proportional hazards model as loss function. Due to the generally limited number of observations in clinical studies, combining different data sets has been proposed to improve learning of network parameters. However, if baseline hazards differ between the studies, the assumptions of Cox proportional hazards model are violated. Based on high dimensional transcriptome profiles from different tumor entities, we demonstrate how using a stratified partial likelihood as loss function allows for accounting for the different baseline hazards in a deep learning framework. Additionally, we compare the partial likelihood with the ranking loss, which is frequently employed as loss function in machine learning approaches due to its seemingly simplicity. Using RNA-seq data from the Cancer Genome Atlas (TCGA) we show that use of stratified loss functions leads to an overall better discriminatory power and lower prediction error compared to their non-stratified counterparts. We investigate which genes are identified to have the greatest marginal impact on prediction of survival when using different loss functions. We find that while similar genes are identified, in particular known prognostic genes receive higher importance from stratified loss functions. Taken together, pooling data from different sources for improved parameter learning of deep neural networks benefits largely from employing stratified loss functions that consider potentially varying baseline hazards. For easy application, we provide PyTorch code for stratified loss functions and an explanatory Jupyter notebook in a GitHub repository.

Gene expression variability in long-term survivors of childhood cancer and cancer-free controls in response to ionizing irradiation.

BACKGROUND: Differential expression analysis is usually adjusted for variation. However, most studies that examined the expression variability (EV) have used computations affected by low expression levels and did not examine healthy tissue. This study aims to calculate and characterize an unbiased EV in primary fibroblasts of childhood cancer survivors and cancer-free controls (N0) in response to ionizing radiation. METHODS: Human skin fibroblasts of 52 donors with a first primary neoplasm in childhood (N1), 52 donors with at least one second primary neoplasm (N2 +), as well as 52 N0 were obtained from the KiKme case-control study and exposed to a high (2 Gray) and a low dose (0.05 Gray) of X-rays and sham- irradiation (0 Gray). Genes were then classified as hypo-, non-, or hyper-variable per donor group and radiation treatment, and then examined for over-represented functional signatures. RESULTS: We found 22 genes with considerable EV differences between donor groups, of which 11 genes were associated with response to ionizing radiation, stress, and DNA repair. The largest number of genes exclusive to one donor group and variability classification combination were all detected in N0: hypo-variable genes after 0 Gray (n = 49), 0.05 Gray (n = 41), and 2 Gray (n = 38), as well as hyper-variable genes after any dose (n = 43). While after 2 Gray positive regulation of cell cycle was hypo-variable in N0, (regulation of) fibroblast proliferation was over-represented in hyper-variable genes of N1 and N2+. In N2+, 30 genes were uniquely classified as hyper-variable after the low dose and were associated with the ERK1/ERK2 cascade. For N1, no exclusive gene sets with functions related to the radiation response were detected in our data. CONCLUSION: N2+ showed high degrees of variability in pathways for the cell fate decision after genotoxic insults that may lead to the transfer and multiplication of DNA-damage via proliferation, where apoptosis and removal of the damaged genome would have been appropriate. Such a deficiency could potentially lead to a higher vulnerability towards side effects of exposure to high doses of ionizing radiation, but following low-dose applications employed in diagnostics, as well.

Synthetic observations from deep generative models and binary omics data with limited sample size.

Deep generative models can be trained to represent the joint distribution of data, such as measurements of single nucleotide polymorphisms (SNPs) from several individuals. Subsequently, synthetic observations are obtained by drawing from this distribution. This has been shown to be useful for several tasks, such as removal of noise, imputation, for better understanding underlying patterns, or even exchanging data under privacy constraints. Yet, it is still unclear how well these approaches work with limited sample size. We investigate such settings specifically for binary data, e.g. as relevant when considering SNP measurements, and evaluate three frequently employed generative modeling approaches, variational autoencoders (VAEs), deep Boltzmann machines (DBMs) and generative adversarial networks (GANs). This includes conditional approaches, such as when considering gene expression conditional on SNPs. Recovery of pair-wise odds ratios (ORs) is considered as a primary performance criterion. For simulated as well as real SNP data, we observe that DBMs generally can recover structure for up to 300 variables, with a tendency of over-estimating ORs when not carefully tuned. VAEs generally get the direction and relative strength of pairwise relations right, yet with considerable under-estimation of ORs. GANs provide stable results only with larger sample sizes and strong pair-wise relations in the data. Taken together, DBMs and VAEs (in contrast to GANs) appear to be well suited for binary omics data, even at rather small sample sizes. This opens the way for many potential applications where synthetic observations from omics data might be useful.

Epidemiology of inherited epidermolysis bullosa in Germany.

BACKGROUND: Epidermolysis bullosa (EB) is a rare genetic disorder manifesting with skin and mucosal membrane blistering in different degrees of severity. OBJECTIVE: Epidemiological data from different countries have been published, but none are available from Germany. METHODS: In this population-based cross-sectional study, people living with EB in Germany were identified using the following sources: academic hospitals, diagnostic laboratories and patient organization. RESULTS: Our study indicates an overall EB incidence of 45 per million live births in Germany. With 14.23 per million live births for junctional EB, the incidence is higher than in other countries, possibly reflecting the availability of early molecular genetic diagnostics in severely affected neonates. Dystrophic EB was assessed at 15.58 cases per million live births. The relatively low incidence found for EB simplex, 14.93 per million live births, could be explained by late or missed diagnosis, but also by 33% of cases remaining not otherwise specified. Using log-linear models, we estimated a prevalence of 54 per million for all EB types, 2.44 for junctional EB, 12.16 for dystrophic EB and 28.44 per million for EB simplex. These figures are comparable to previously reported data from other countries. CONCLUSIONS: Altogether, there are at least 2000 patients with EB in the German population. These results should support national policies and pharmaceutical companies in decision-making, allow more precise planning of drug development and clinical trials, and aid patient advocacy groups in their effort to improve quality of life of people with this orphan disease.

[The situation of informal caregivers of working age in the COVID-19 pandemic-results of an online survey in Germany]. / Die Situation pflegender Angehöriger im erwerbsfähigen Alter in der COVID-19-Pandemie ­ Ergebnisse einer Onlinebefragung in Deutschland.

BACKGROUND: The consequences of the COVID-19 pandemic have posed major challenges to different groups. One of these are informal caregivers. This study investigates the changes the pandemic has caused for informal caregivers and the extent to which quality of life and burden of care have changed for specific subgroups. METHODS: Data for this cross-sectional study was gathered in the summer of 2020 in a convenient sample of informal caregivers (< 67 years of age, N = 1143). In addition to sociodemographic data, information on the care situation, compatibility of care and work, as well as stress and quality of life was collected in an online survey. The analysis of care situations and compatibility of care and work is done descriptively. Logistic regression models are used for a subgroup analysis of quality of life and care burden. RESULTS: The care situation has changed for 54.7% of participants and has become more time consuming. For 70.8% of respondents, the COVID-19 pandemic has made it even more difficult to balance care-giving and work. However, most respondents were satisfied with their employers' pandemic management (65.9%). A sharp decline in the quality of life and an increase in the burden of care for informal caregivers was ascertained. Both developments are stronger for young and female caregivers and for those caring for people with a greater need of support. DISCUSSION: The results indicate that living situations worsened for a substantial proportion of informal caregivers during the COVID-19 pandemic. Policymakers should recognize additional challenges that informal caregivers have faced since the outbreak of the COVID-19 pandemic and how they vary by subgroups. It is important to include home-based informal care as well as other care settings in future pandemic concepts.

T-cell exhaustion and residency dynamics inform clinical outcomes in hepatocellular carcinoma.

BACKGROUND & AIMS: Despite recent translation of immunotherapies into clinical practice, the immunobiology of hepatocellular carcinoma (HCC), in particular the role and clinical relevance of exhausted and liver-resident T cells remain unclear. We therefore dissected the landscape of exhausted and resident T cell responses in the peripheral blood and tumor microenvironment of patients with HCC. METHODS: Lymphocytes were isolated from the blood, tumor and tumor-surrounding liver tissue of patients with HCC (n = 40, n = 10 treated with anti-PD-1 therapy). Phenotype, function and response to anti-PD-1 were analyzed by mass and flow cytometry ex vivo and in vitro, tissue residence was further assessed by immunohistochemistry and imaging mass cytometry. Gene signatures were analyzed in silico. RESULTS: We identified significant enrichment of heterogeneous populations of exhausted CD8+ T cells (TEX) in the tumor microenvironment. Strong enrichment of severely exhausted CD8 T cells expressing multiple immune checkpoints in addition to PD-1 was linked to poor progression-free and overall survival. In contrast, PD-1 was also expressed on a subset of more functional and metabolically active CD103+ tissue-resident memory T cells (TRM) that expressed few additional immune checkpoints and were associated with better survival. TEX enrichment was independent of BCLC stage, alpha-fetoprotein levels or age as a variable for progression-free survival in our cohort. These findings were in line with in silico gene signature analysis of HCC tumor transcriptomes from The Cancer Genome Atlas. A higher baseline TRM/TEX ratio was associated with disease control in anti-PD-1-treated patients. CONCLUSION: Our data provide information on the role of peripheral and intratumoral TEX-TRM dynamics in determining outcomes in patients with HCC. The dynamics between exhausted and liver-resident T cells have implications for immune-based diagnostics, rational patient selection and monitoring during HCC immunotherapies. LAY SUMMARY: The role of the immune response in hepatocellular carcinoma (HCC) remains unclear. T cells can mediate protection against tumor cells but are frequently dysfunctional and exhausted in cancer. We found that patients with a predominance of exhausted CD8+ T cells (TEX) had poor survival compared to patients with a predominance of tissue-resident memory T cells (TRM). This correlated with the molecular profile, metabolic and functional status of these cell populations. The enrichment of TEX was independently associated with prognosis in addition to disease stage, age and tumor markers. A high TRM proportion was also associated with better outcomes following checkpoint therapy. Thus, these T-cell populations are novel biomarkers with relevance in HCC.

Interpretable generative deep learning: an illustration with single cell gene expression data.

Deep generative models can learn the underlying structure, such as pathways or gene programs, from omics data. We provide an introduction as well as an overview of such techniques, specifically illustrating their use with single-cell gene expression data. For example, the low dimensional latent representations offered by various approaches, such as variational auto-encoders, are useful to get a better understanding of the relations between observed gene expressions and experimental factors or phenotypes. Furthermore, by providing a generative model for the latent and observed variables, deep generative models can generate synthetic observations, which allow us to assess the uncertainty in the learned representations. While deep generative models are useful to learn the structure of high-dimensional omics data by efficiently capturing non-linear dependencies between genes, they are sometimes difficult to interpret due to their neural network building blocks. More precisely, to understand the relationship between learned latent variables and observed variables, e.g., gene transcript abundances and external phenotypes, is difficult. Therefore, we also illustrate current approaches that allow us to infer the relationship between learned latent variables and observed variables as well as external phenotypes. Thereby, we render deep learning approaches more interpretable. In an application with single-cell gene expression data, we demonstrate the utility of the discussed methods.

Radiation-response in primary fibroblasts of long-term survivors of childhood cancer with and without second primary neoplasms: the KiKme study.

BACKGROUND: The etiology and most risk factors for a sporadic first primary neoplasm in childhood or subsequent second primary neoplasms are still unknown. One established causal factor for therapy-associated second primary neoplasms is the exposure to ionizing radiation during radiation therapy as a mainstay of cancer treatment. Second primary neoplasms occur in 8% of all cancer survivors within 30 years after the first diagnosis in Germany, but the underlying factors for intrinsic susceptibilities have not yet been clarified. Thus, the purpose of this nested case-control study was the investigation and comparison of gene expression and affected pathways in primary fibroblasts of childhood cancer survivors with a first primary neoplasm only or with at least one subsequent second primary neoplasm, and controls without neoplasms after exposure to a low and a high dose of ionizing radiation. METHODS: Primary fibroblasts were obtained from skin biopsies from 52 adult donors with a first primary neoplasm in childhood (N1), 52 with at least one additional primary neoplasm (N2+), as well as 52 without cancer (N0) from the KiKme study. Cultured fibroblasts were exposed to a high [2 Gray (Gy)] and a low dose (0.05 Gy) of X-rays. Messenger ribonucleic acid was extracted 4 h after exposure and Illumina-sequenced. Differentially expressed genes (DEGs) were computed using limma for R, selected at a false discovery rate level of 0.05, and further analyzed for pathway enrichment (right-tailed Fisher's Exact Test) and (in-) activation (z ≥|2|) using Ingenuity Pathway Analysis. RESULTS: After 0.05 Gy, least DEGs were found in N0 (n = 236), compared to N1 (n = 653) and N2+ (n = 694). The top DEGs with regard to the adjusted p-value were upregulated in fibroblasts across all donor groups (SESN1, MDM2, CDKN1A, TIGAR, BTG2, BLOC1S2, PPM1D, PHLDB3, FBXO22, AEN, TRIAP1, and POLH). Here, we observed activation of p53 Signaling in N0 and to a lesser extent in N1, but not in N2+. Only in N0, DNA (excision-) repair (involved genes: CDKN1A, PPM1D, and DDB2) was predicted to be a downstream function, while molecular networks in N2+ were associated with cancer, as well as injury and abnormalities (among others, downregulation of MSH6, CCNE2, and CHUK). After 2 Gy, the number of DEGs was similar in fibroblasts of all donor groups and genes with the highest absolute log2 fold-change were upregulated throughout (CDKN1A, TIGAR, HSPA4L, MDM2, BLOC1SD2, PPM1D, SESN1, BTG2, FBXO22, PCNA, and TRIAP1). Here, the p53 Signaling-Pathway was activated in fibroblasts of all donor groups. The Mitotic Roles of Polo Like Kinase-Pathway was inactivated in N1 and N2+. Molecular Mechanisms of Cancer were affected in fibroblasts of all donor groups. P53 was predicted to be an upstream regulator in fibroblasts of all donor groups and E2F1 in N1 and N2+. Results of the downstream analysis were senescence in N0 and N2+, transformation of cells in N0, and no significant effects in N1. Seven genes were differentially expressed in reaction to 2 Gy dependent on the donor group (LINC00601, COBLL1, SESN2, BIN3, TNFRSF10A, EEF1AKNMT, and BTG2). CONCLUSION: Our results show dose-dependent differences in the radiation response between N1/N2+ and N0. While mechanisms against genotoxic stress were activated to the same extent after a high dose in all groups, the radiation response was impaired after a low dose in N1/N2+, suggesting an increased risk for adverse effects including carcinogenesis, particularly in N2+.

Exploring generative deep learning for omics data using log-linear models.

MOTIVATION: Following many successful applications to image data, deep learning is now also increasingly considered for omics data. In particular, generative deep learning not only provides competitive prediction performance, but also allows for uncovering structure by generating synthetic samples. However, exploration and visualization is not as straightforward as with image applications. RESULTS: We demonstrate how log-linear models, fitted to the generated, synthetic data can be used to extract patterns from omics data, learned by deep generative techniques. Specifically, interactions between latent representations learned by the approaches and generated synthetic data are used to determine sets of joint patterns. Distances of patterns with respect to the distribution of latent representations are then visualized in low-dimensional coordinate systems, e.g. for monitoring training progress. This is illustrated with simulated data and subsequently with cortical single-cell gene expression data. Using different kinds of deep generative techniques, specifically variational autoencoders and deep Boltzmann machines, the proposed approach highlights how the techniques uncover underlying structure. It facilitates the real-world use of such generative deep learning techniques to gain biological insights from omics data. AVAILABILITY AND IMPLEMENTATION: The code for the approach as well as an accompanying Jupyter notebook, which illustrates the application of our approach, is available via the GitHub repository: https://github.com/ssehztirom/Exploring-generative-deep-learning-for-omics-data-by-using-log-linear-models. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Deep generative models in DataSHIELD.

BACKGROUND: The best way to calculate statistics from medical data is to use the data of individual patients. In some settings, this data is difficult to obtain due to privacy restrictions. In Germany, for example, it is not possible to pool routine data from different hospitals for research purposes without the consent of the patients. METHODS: The DataSHIELD software provides an infrastructure and a set of statistical methods for joint, privacy-preserving analyses of distributed data. The contained algorithms are reformulated to work with aggregated data from the participating sites instead of the individual data. If a desired algorithm is not implemented in DataSHIELD or cannot be reformulated in such a way, using artificial data is an alternative. Generating artificial data is possible using so-called generative models, which are able to capture the distribution of given data. Here, we employ deep Boltzmann machines (DBMs) as generative models. For the implementation, we use the package "BoltzmannMachines" from the Julia programming language and wrap it for use with DataSHIELD, which is based on R. RESULTS: We present a methodology together with a software implementation that builds on DataSHIELD to create artificial data that preserve complex patterns from distributed individual patient data. Such data sets of artificial patients, which are not linked to real patients, can then be used for joint analyses. As an exemplary application, we conduct a distributed analysis with DBMs on a synthetic data set, which simulates genetic variant data. Patterns from the original data can be recovered in the artificial data using hierarchical clustering of the virtual patients, demonstrating the feasibility of the approach. Additionally, we compare DBMs, variational autoencoders, generative adversarial networks, and multivariate imputation as generative approaches by assessing the utility and disclosure of synthetic data generated from real genetic variant data in a distributed setting with data of a small sample size. CONCLUSIONS: Our implementation adds to DataSHIELD the ability to generate artificial data that can be used for various analyses, e.g., for pattern recognition with deep learning. This also demonstrates more generally how DataSHIELD can be flexibly extended with advanced algorithms from languages other than R.

Comparison of time and dose dependent gene expression and affected pathways in primary human fibroblasts after exposure to ionizing radiation.

BACKGROUND: Exposure to ionizing radiation induces complex stress responses in cells, which can lead to adverse health effects such as cancer. Although a variety of studies investigated gene expression and affected pathways in human fibroblasts after exposure to ionizing radiation, the understanding of underlying mechanisms and biological effects is still incomplete due to different experimental settings and small sample sizes. Therefore, this study aims to identify the time point with the highest number of differentially expressed genes and corresponding pathways in primary human fibroblasts after irradiation at two preselected time points. METHODS: Fibroblasts from skin biopsies of 15 cell donors were exposed to a high (2Gy) and a low (0.05Gy) dose of X-rays. RNA was extracted and sequenced 2 h and 4 h after exposure. Differentially expressed genes with an adjusted p-value < 0.05 were flagged and used for pathway analyses including prediction of upstream and downstream effects. Principal component analyses were used to examine the effect of two different sequencing runs on quality metrics and variation in expression and alignment and for explorative analysis of the radiation dose and time point of analysis. RESULTS: More genes were differentially expressed 4 h after exposure to low and high doses of radiation than after 2 h. In experiments with high dose irradiation and RNA sequencing after 4 h, inactivation of the FAT10 cancer signaling pathway and activation of gluconeogenesis I, glycolysis I, and prostanoid biosynthesis was observed taking p-value (< 0.05) and (in) activating z-score (≥2.00 or ≤ - 2.00) into account. Two hours after high dose irradiation, inactivation of small cell lung cancer signaling was observed. For low dose irradiation experiments, we did not detect any significant (p < 0.05 and z-score ≥ 2.00 or ≤ - 2.00) activated or inactivated pathways for both time points. CONCLUSIONS: Compared to 2 h after irradiation, a higher number of differentially expressed genes were found 4 h after exposure to low and high dose ionizing radiation. Differences in gene expression were related to signal transduction pathways of the DNA damage response after 2 h and to metabolic pathways, that might implicate cellular senescence, after 4 h. The time point 4 h will be used to conduct further irradiation experiments in a larger sample.

Educational differences in the influence of health on early work exit among older workers.

OBJECTIVES: Previous research has shown that poor physical and mental health are important risk factors for early work exit. We examined potential differences in this association in older workers (50+) across educational levels. METHODS: Coordinated analyses were carried out in longitudinal data sets from four European countries: the Netherlands (Longitudinal Aging Study Amsterdam), Denmark (Danish Longitudinal Study of Ageing), England (English Longitudinal Study of Ageing) and Germany (German Ageing Survey). The effect of poor self-rated health (SRH), functional limitations and depression on different types of early work exit (early retirement, economic inactivity, disability and unemployment) was examined using Cox regression analysis. We examined educational differences in these effects by testing interaction terms. RESULTS: Poor physical and mental health were more common among the lower educated. Poor SRH, functional limitations, and depression were all associated with a higher risk of early work exit. These health effects were strongest for the disability exit routes (poor SRH: HRs 5.77 to 8.14; functional limitations: HRs 6.65 to 10.42; depression: HRs 3.30 to 5.56). In the Netherlands (functional limitations) and England (functional limitations and SRH), effects were stronger in the lower educated. CONCLUSIONS: The prevalence of health problems, that is, poor SRH, functional limitations and depression, was higher in the lower educated workers. All three health indicators increase the risk of early work exit. In some countries, health effects on early exit were stronger in the lower educated. Thus, lower educated older workers are an important target group for health policy and intervention.

BAL Cell Gene Expression Is Indicative of Outcome and Airway Basal Cell Involvement in Idiopathic Pulmonary Fibrosis.

RATIONALE: Idiopathic pulmonary fibrosis (IPF) is a fatal disease with a variable and unpredictable course. OBJECTIVES: To determine whether BAL cell gene expression is predictive of survival in IPF. METHODS: This retrospective study analyzed the BAL transcriptome of three independent IPF cohorts: Freiburg (Germany), Siena (Italy), and Leuven (Belgium) including 212 patients. BAL cells from 20 healthy volunteers, 26 patients with sarcoidosis stage III and IV, and 29 patients with chronic obstructive pulmonary disease were used as control subjects. Survival analysis was performed by Cox models and component-wise boosting. Presence of airway basal cells was tested by immunohistochemistry and flow cytometry. MEASUREMENTS AND MAIN RESULTS: A total of 1,582 genes were predictive of mortality in the IPF derivation cohort in univariate analyses adjusted for age and sex at false discovery rate less than 0.05. A nine-gene signature, derived from the discovery cohort (Freiburg), performed well in both replication cohorts, Siena (P < 0.0032) and Leuven (P = 0.0033). nCounter expression analysis confirmed the array results (P < 0.0001). The genes associated with mortality in BAL cells were significantly enriched for genes expressed in airway basal cells. Further analyses by gene expression, flow cytometry, and immunohistochemistry showed an increase in airway basal cells in BAL and tissues of IPF compared with control subjects, but not in chronic obstructive pulmonary disease or sarcoidosis. CONCLUSIONS: Our results identify and validate a BAL signature that predicts mortality in IPF and improves the accuracy of outcome prediction based on clinical parameters. The BAL signature associated with mortality unmasks a potential role for airway basal cells in IPF.

[Change of retirement transition in the political paradigm shift in Europe : Late career and retirement transitions in Germany, Austria, Sweden and Estonia]. / Wandel von Ruhestandsübergängen im politischen Paradigmenwechsel Europas : Späte Erwerbsphase und Rentenübergang in Deutschland, Österreich, Schweden und Estland.

BACKGROUND: In the last two decades labor market participation for older employees has undergone a gradual political paradigm shift in many European countries from a policy of early retirement to one of extending working lives and active aging. OBJECTIVE: This study investigated if this political paradigm shift is causing new social inequalities in retirement transition due to restricted financial possibilities for early retirement. MATERIAL AND METHODS: Data were derived from the European Union Labor Force Survey from the years 2006 and 2012 and selected European countries (Germany, Austria, Sweden and Estonia) were analyzed. RESULTS: Associations between the specific implementation of the policy of active aging, the freedom of choice in retirement timing and retirement transition were found. It seems that voluntary retirement transitions are highest in those countries where the labor market and social policies are most coherent and aimed at supporting older workers' employability. CONCLUSION: The reduction of early retirement incentives should be supported by active labor market policies and a policy of extensive age-independent further training measures in order to minimize social inequalities.

Transition to retirement in the healthcare sector : Working conditions and attitudes of older workers.

BACKGROUND: The aging of societies will increase the need for healthcare services and lead to a growing number of older employees. These two developments are relevant in the healthcare sector (HCS), which is of rising societal and economic importance and at the same time employs many older people. OBJECTIVE: This article, which was written within the EXTEND project, investigates the working conditions and the prospective retirement age of older employees in the HCS in comparison to other sectors and explores what companies are doing to address the needs of this group. MATERIAL AND METHODS: The analysis was conducted as a mixed methods approach. The quantitative part was based on data derived from the Transitions and Old Age Potential (TOP) study in which older German employees were asked about their working conditions and retirement transitions. Matching techniques (coarsened exact matching) were used to investigate differences between sectors. The sample consisted of 114 employees aged between 55 and 65 years, working in the HCS and their statistical twins. The qualitative analysis was based on case studies in two inpatient care organizations and two hospitals in Germany. A total of 23 semistructured interviews with staff members and with representatives of the management were carried out and thematically analyzed. RESULTS: The results showed that older employees in the HCS do not expect to retire earlier but preferred to do so significantly more often. Furthermore, HCS employees are more likely to face physically burdensome working conditions than in other sectors of the economy. The case studies indicated that there are very diverse and unsystematic strategies in addressing and supporting older employees. DISCUSSION: Older employees in the HCS sector are employed in much harsher working conditions than their peers in other sectors. This must be kept in mind when trying to extend their working lives.

Systematically evaluating interfaces for RNA-seq analysis from a life scientist perspective.

RNA-sequencing (RNA-seq) has become an established way for measuring gene expression in model organisms and humans. While methods development for refining the corresponding data processing and analysis pipeline is ongoing, protocols for typical steps have been proposed and are widely used. Several user interfaces have been developed for making such analysis steps accessible to life scientists without extensive knowledge of command line tools. We performed a systematic search and evaluation of such interfaces to investigate to what extent these can indeed facilitate RNA-seq data analysis. We found a total of 29 open source interfaces, and six of the more widely used interfaces were evaluated in detail. Central criteria for evaluation were ease of configuration, documentation, usability, computational demand and reporting. No interface scored best in all of these criteria, indicating that the final choice will depend on the specific perspective of users and the corresponding weighting of criteria. Considerable technical hurdles had to be overcome in our evaluation. For many users, this will diminish potential benefits compared with command line tools, leaving room for future improvement of interfaces.

Partitioned learning of deep Boltzmann machines for SNP data.

MOTIVATION: Learning the joint distributions of measurements, and in particular identification of an appropriate low-dimensional manifold, has been found to be a powerful ingredient of deep leaning approaches. Yet, such approaches have hardly been applied to single nucleotide polymorphism (SNP) data, probably due to the high number of features typically exceeding the number of studied individuals. RESULTS: After a brief overview of how deep Boltzmann machines (DBMs), a deep learning approach, can be adapted to SNP data in principle, we specifically present a way to alleviate the dimensionality problem by partitioned learning. We propose a sparse regression approach to coarsely screen the joint distribution of SNPs, followed by training several DBMs on SNP partitions that were identified by the screening. Aggregate features representing SNP patterns and the corresponding SNPs are extracted from the DBMs by a combination of statistical tests and sparse regression. In simulated case-control data, we show how this can uncover complex SNP patterns and augment results from univariate approaches, while maintaining type 1 error control. Time-to-event endpoints are considered in an application with acute myeloid leukemia patients, where SNP patterns are modeled after a pre-screening based on gene expression data. The proposed approach identified three SNPs that seem to jointly influence survival in a validation dataset. This indicates the added value of jointly investigating SNPs compared to standard univariate analyses and makes partitioned learning of DBMs an interesting complementary approach when analyzing SNP data. AVAILABILITY AND IMPLEMENTATION: A Julia package is provided at 'http://github.com/binderh/BoltzmannMachines.jl'. CONTACT: binderh@imbi.uni-freiburg.de. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

A genome-wide association study of corneal astigmatism: The CREAM Consortium.

Purpose: To identify genes and genetic markers associated with corneal astigmatism. Methods: A meta-analysis of genome-wide association studies (GWASs) of corneal astigmatism undertaken for 14 European ancestry (n=22,250) and 8 Asian ancestry (n=9,120) cohorts was performed by the Consortium for Refractive Error and Myopia. Cases were defined as having >0.75 diopters of corneal astigmatism. Subsequent gene-based and gene-set analyses of the meta-analyzed results of European ancestry cohorts were performed using VEGAS2 and MAGMA software. Additionally, estimates of single nucleotide polymorphism (SNP)-based heritability for corneal and refractive astigmatism and the spherical equivalent were calculated for Europeans using LD score regression. Results: The meta-analysis of all cohorts identified a genome-wide significant locus near the platelet-derived growth factor receptor alpha (PDGFRA) gene: top SNP: rs7673984, odds ratio=1.12 (95% CI:1.08-1.16), p=5.55×10-9. No other genome-wide significant loci were identified in the combined analysis or European/Asian ancestry-specific analyses. Gene-based analysis identified three novel candidate genes for corneal astigmatism in Europeans-claudin-7 (CLDN7), acid phosphatase 2, lysosomal (ACP2), and TNF alpha-induced protein 8 like 3 (TNFAIP8L3). Conclusions: In addition to replicating a previously identified genome-wide significant locus for corneal astigmatism near the PDGFRA gene, gene-based analysis identified three novel candidate genes, CLDN7, ACP2, and TNFAIP8L3, that warrant further investigation to understand their role in the pathogenesis of corneal astigmatism. The much lower number of genetic variants and genes demonstrating an association with corneal astigmatism compared to published spherical equivalent GWAS analyses suggest a greater influence of rare genetic variants, non-additive genetic effects, or environmental factors in the development of astigmatism.

Expected and preferred retirement age in Germany.

BACKGROUND: Over the last 10 years the German pension system has undergone several reforms including the abandonment of early retirement policies and an increase in the statutory retirement age. Consequently, the average retirement age has increased and future retiree cohorts have adjusted the retirement expectations and preferences as to when they would like to retire. OBJECTIVE: This study was carried out to examine discrepancies between the expected and the preferred retirement age of older workers in Germany and to investigate how these discrepancies differ between groups of older workers. MATERIAL AND METHODS: Based on data from the survey "Employment after retirement", the expected and preferred retirement ages of 1500 workers aged 55 years and older were compared. Regression analyses were used to investigate the influence of educational level and professional position on deviances between the expected and preferred retirement ages. RESULTS: On average older workers would like to retire 1.75 years earlier than they actually expect to. The deviance is significantly larger for employees with a lower professional position, lower income and lower educational level. CONCLUSION: The discrepancy between expected and preferred retirement ages, in particular for older workers in vulnerable labor market positions, indicates a potential social inequality regarding the choice of retirement timing. This must be acknowledged when considering further reforms of the German pension system.

Working Hours Flexibility and Timing of Retirement: Findings from Europe.

This study investigates how flexibility in working hours affects retirement timing. It tests the assumption that decreasing weekly working hours delays retirement and extends working life. Using data from four waves of the Survey of Health, Ageing and Retirement in Europe (SHARE) and of the English Longitudinal Study of Ageing (ELSA), we analyze whether a shift from full-time to part-time work delays retirement. Results show that older workers who reduce their working hours retire earlier than those who stay in full-time employment. The effect is stronger in Central and Eastern Europe than in Scandinavian countries. No interaction effects for gender and work strain are found. We conclude that part-time work at the end of the career, as a means to extend working life, should be reevaluated.