ELIXIR is a pan-European intergovernmental organisation for life science that aims to coordinate bioinformatics resources in a single infrastructure across Europe; bioinformatics training is central to its strategy, which aims to develop a training community that spans all ELIXIR member states. In an evidence-based approach for strengthening bioinformatics training programmes across Europe, the ELIXIR Training Platform, led by the ELIXIR EXCELERATE Quality and Impact Assessment Subtask in collaboration with the ELIXIR Training Coordinators Group, has implemented an assessment strategy to measure quality and impact of its entire training portfolio. Here, we present ELIXIR's framework for assessing training quality and impact, which includes the following: specifying assessment aims, determining what data to collect in order to address these aims, and our strategy for centralised data collection to allow for ELIXIR-wide analyses. In addition, we present an overview of the ELIXIR training data collected over the past 4 years. We highlight the importance of a coordinated and consistent data collection approach and the relevance of defining specific metrics and answer scales for consortium-wide analyses as well as for comparison of data across iterations of the same course.

BACKGROUND: It is of great importance to identify molecular processes and pathways that are involved in disease etiology. Although there has been an extensive use of various high-throughput methods for this task, pathogenic pathways are still not completely understood. Often the set of genes or proteins identified as altered in genome-wide screens show a poor overlap with canonical disease pathways. These findings are difficult to interpret, yet crucial in order to improve the understanding of the molecular processes underlying the disease progression. We present a novel method for identifying groups of connected molecules from a set of differentially expressed genes. These groups represent functional modules sharing common cellular function and involve signaling and regulatory events. Specifically, our method makes use of Bayesian statistics to identify groups of co-regulated genes based on the microarray data, where external information about molecular interactions and connections are used as priors in the group assignments. Markov chain Monte Carlo sampling is used to search for the most reliable grouping. RESULTS: Simulation results showed that the method improved the ability of identifying correct groups compared to traditional clustering, especially for small sample sizes. Applied to a microarray heart failure dataset the method found one large cluster with several genes important for the structure of the extracellular matrix and a smaller group with many genes involved in carbohydrate metabolism. The method was also applied to a microarray dataset on melanoma cancer patients with or without metastasis, where the main cluster was dominated by genes related to keratinocyte differentiation. CONCLUSION: Our method found clusters overlapping with known pathogenic processes, but also pointed to new connections extending beyond the classical pathways.

MOTIVATION: In molecular biology, as in many other scientific fields, the scale of analyses is ever increasing. Often, complex Monte Carlo simulation is required, sometimes within a large-scale multiple testing setting. The resulting computational costs may be prohibitively high. RESULTS: We here present MCFDR, a simple, novel algorithm for false discovery rate (FDR) modulated sequential Monte Carlo (MC) multiple hypothesis testing. The algorithm iterates between adding MC samples across tests and calculating intermediate FDR values for the collection of tests. MC sampling is stopped either by sequential MC or based on a threshold on FDR. An essential property of the algorithm is that it limits the total number of MC samples whatever the number of true null hypotheses. We show on both real and simulated data that the proposed algorithm provides large gains in computational efficiency. AVAILABILITY: MCFDR is implemented in the Genomic HyperBrowser (http://hyperbrowser.uio.no/mcfdr), a web-based system for genome analysis. All input data and results are available and can be reproduced through a Galaxy Pages document at: http://hyperbrowser.uio.no/mcfdr/u/sandve/p/mcfdr. CONTACT: [email protected].

BACKGROUND: Survival prediction from high-dimensional genomic data is an active field in today's medical research. Most of the proposed prediction methods make use of genomic data alone without considering established clinical covariates that often are available and known to have predictive value. Recent studies suggest that combining clinical and genomic information may improve predictions, but there is a lack of systematic studies on the topic. Also, for the widely used Cox regression model, it is not obvious how to handle such combined models. RESULTS: We propose a way to combine classical clinical covariates with genomic data in a clinico-genomic prediction model based on the Cox regression model. The prediction model is obtained by a simultaneous use of both types of covariates, but applying dimension reduction only to the high-dimensional genomic variables. We describe how this can be done for seven well-known prediction methods: variable selection, unsupervised and supervised principal components regression and partial least squares regression, ridge regression, and the lasso. We further perform a systematic comparison of the performance of prediction models using clinical covariates only, genomic data only, or a combination of the two. The comparison is done using three survival data sets containing both clinical information and microarray gene expression data. Matlab code for the clinico-genomic prediction methods is available at http://www.med.uio.no/imb/stat/bmms/software/clinico-genomic/. CONCLUSIONS: Based on our three data sets, the comparison shows that established clinical covariates will often lead to better predictions than what can be obtained from genomic data alone. In the cases where the genomic models are better than the clinical, ridge regression is used for dimension reduction. We also find that the clinico-genomic models tend to outperform the models based on only genomic data. Further, clinico-genomic models and the use of ridge regression gives for all three data sets better predictions than models based on the clinical covariates alone.

MOTIVATION: Survival prediction from gene expression data and other high-dimensional genomic data has been subject to much research during the last years. These kinds of data are associated with the methodological problem of having many more gene expression values than individuals. In addition, the responses are censored survival times. Most of the proposed methods handle this by using Cox's proportional hazards model and obtain parameter estimates by some dimension reduction or parameter shrinkage estimation technique. Using three well-known microarray gene expression data sets, we compare the prediction performance of seven such methods: univariate selection, forward stepwise selection, principal components regression (PCR), supervised principal components regression, partial least squares regression (PLS), ridge regression and the lasso. RESULTS: Statistical learning from subsets should be repeated several times in order to get a fair comparison between methods. Methods using coefficient shrinkage or linear combinations of the gene expression values have much better performance than the simple variable selection methods. For our data sets, ridge regression has the overall best performance. AVAILABILITY: Matlab and R code for the prediction methods are available at http://www.med.uio.no/imb/stat/bmms/software/microsurv/.