Marcel Dettling

Bioconductor: open software development for computational biology and bioinformatics

The Bioconductor project is an initiative for the collaborative creation of extensible software for computational biology and bioinformatics. The goals of the project include: fostering collaborative development and widespread use of innovative software, reducing barriers to entry into interdisciplinary scientific research, and promoting the achievement of remote reproducibility of research results. We describe details of our aims and methods, identify current challenges, compare Bioconductor to other open bioinformatics projects, and provide working examples.

BagBoosting for tumor classification with gene expression data

MOTIVATION Microarray experiments are expected to contribute significantly to the progress in cancer treatment by enabling a precise and early diagnosis. They create a need for class prediction tools, which can deal with a large number of highly correlated input variables, perform feature selection and provide class probability estimates that serve as a quantification of the predictive uncertainty. A very promising solution is to combine the two ensemble schemes bagging and boosting to a novel algorithm called BagBoosting. RESULTS When bagging is used as a module in boosting, the resulting classifier consistently improves the predictive performance and the probability estimates of both bagging and boosting on real and simulated gene expression data. This quasi-guaranteed improvement can be obtained by simply making a bigger computing effort. The advantageous predictive potential is also confirmed by comparing BagBoosting to several established class prediction tools for microarray data. AVAILABILITY Software for the modified boosting algorithms, for benchmark studies and for the simulation of microarray data are available as an R package under GNU public license at http://stat.ethz.ch/~dettling/bagboost.html.

Gene Expression Signatures Identify Rhabdomyosarcoma Subtypes and Detect a Novel t(2;2)(q35;p23) Translocation Fusing PAX3 to NCOA1

Rhabdomyosarcoma is a pediatric tumor type, which is classified based on histological criteria into two major subgroups, namely embryonal rhabdomyosarcoma and alveolar rhabdomyosarcoma. The majority, but not all, alveolar rhabdomyosarcoma carry the specific PAX3(7)/FKHR-translocation, whereas there is no consistent genetic abnormality recognized in embryonal rhabdomyosarcoma. To gain additional insight into the genetic characteristics of these subtypes, we used oligonucleotide microarrays to measure the expression profiles of a group of 29 rhabdomyosarcoma biopsy samples (15 embryonal rhabdomyosarcoma, and 10 translocation-positive and 4 translocation-negative alveolar rhabdomyosarcoma). Hierarchical clustering revealed expression signatures clearly discriminating all three of the subgroups. Differentially expressed genes included several tyrosine kinases and G protein-coupled receptors, which might be amenable to pharmacological intervention. In addition, the alveolar rhabdomyosarcoma signature was used to classify an additional alveolar rhabdomyosarcoma case lacking any known PAX3 or PAX7 fusion as belonging to the translocation-positive group, leading to the identification of a novel translocation t(2;2)(q35;p23), which generates a fusion protein composed of PAX3 and the nuclear receptor coactivator NCOA1, having similar transactivation properties as PAX3/FKHR. These experiments demonstrate for the first time that gene expression profiling is capable of identifying novel chromosomal translocations.

Supervised clustering of genes.

BackgroundWe focus on microarray data where experiments monitor gene expression in different tissues and where each experiment is equipped with an additional response variable such as a cancer type. Although the number of measured genes is in the thousands, it is assumed that only a few marker components of gene subsets determine the type of a tissue. Here we present a new method for finding such groups of genes by directly incorporating the response variables into the grouping process, yielding a supervised clustering algorithm for genes.ResultsAn empirical study on eight publicly available microarray datasets shows that our algorithm identifies gene clusters with excellent predictive potential, often superior to classification with state-of-the-art methods based on single genes. Permutation tests and bootstrapping provide evidence that the output is reasonably stable and more than a noise artifact.ConclusionsIn contrast to other methods such as hierarchical clustering, our algorithm identifies several gene clusters whose expression levels clearly distinguish the different tissue types. The identification of such gene clusters is potentially useful for medical diagnostics and may at the same time reveal insights into functional genomics.

Prenatal origin of separate evolution of leukemia in identical twins.

Several studies involving identical twins with concordant leukemia and retrospective scrutiny of archived neonatal blood spots have shown that the TEL-AML1 fusion gene in childhood acute lymphoblastic leukemia (ALL) frequently arises before birth. A prenatal origin of childhood leukemia was further supported by the detection of clonotypic immunoglobulin gene rearrangements on neonatal blood spots of children with various other subtypes of ALL. However, no comprehensive study is available linking these clonotypic events. We describe a pair of 5-year-old monozygotic twins with concordant TEL-AML1-positive ALL. Separate leukemic clones were identified in the diagnostic samples since distinct IGH and IGK-Kde gene rearrangements could be detected. Additional differences characterizing the leukemic clones included an aberration of the second, nonrearranged TEL allele observed in one twin only. Interestingly, both the identical TEL-AML1 fusion sequence and distinct immunoglobulin gene rearrangements were identified on the neonatal blood spots indicating that separate preleukemic clones evolved already before birth. Finally, we compared the reported twins with an additional 31 children with ALL by using the microarray technology. Gene expression profiling provided evidence that leukemia in twins harbours the same subtype-typical feature as TEL-AML1-positive leukemia in singletons suggesting that the leukemogenesis model might also be applicable generally.

Statistical methods for identifying differentially expressed gene combinations

Identification of coordinate gene expression changes across phenotypes or biological conditions is the basis of the ability to decode the role of gene expression regulatory networks. Statistically, the identification of these changes can be viewed as a search for groups (most typically pairs) of genes whose expression provides better phenotype discrimination when considered jointly than when considered individually. Such groups are defined as being jointly differentially expressed. In this chapter several approaches for identifying jointly differentially expressed groups of genes are reviewed of compared on a set of simulations.

Self-Reported Health Characteristics and Medication Consumption by CAM Users and Nonusers: A Swiss Cross-Sectional Survey

OBJECTIVES Complementary and alternative medicine (CAM) is very popular in Switzerland. The objective of this work was to find out whether the use of CAM therapies is associated with distinct health characteristics and altered consumption of conventional medications. DESIGN AND PARTICIPANTS Self-reported data from the 2007 Swiss Health Survey were analyzed. Two groups of participants were defined and compared with each other: CAM users (those who had used CAM during the last 12 months, n=3333) and nonusers (those who stated they had not used CAM during the last 12 months, n=9821). OUTCOME MEASURES Multivariate logistic regression models were used to determine the predictors of CAM use and to address relevance and magnitude of the differences in medication consumption between CAM users and nonusers. RESULTS Comparatively lower body-mass index (BMI) values and migraine, arthritis, allergies, and depression were associated with increased probability of CAM use. Multivariate logistic regression models that adjusted for the effects of relevant demographic factors, BMI, and perceived health status showed that CAM users consumed fewer medications for cardiovascular diseases--high blood pressure and high cholesterol (and, by trend, heart problems and diabetes)--than nonusers. On the other hand, their consumption of analgesics and medications for depression and for constipation (and, by trend, sedatives and soporifics), was higher than that of nonusers. CONCLUSIONS Migraine, arthritis, depression, and constipation might lead patients to use CAM therapies and, in addition, to consume more of some conventional medications. Given the long intake period and considerable adverse effects of medications, the lower consumption of these agents for chronic cardiovascular problems by CAM users might be beneficial and deserves further investigations.

Volatility and risk estimation with linear and nonlinear methods based on high frequency data

Accurate volatility predictions are crucial for the successful implementation of risk management. The use of high frequency data approximately renders volatility from a latent to an observable quantity, and opens new directions to forecast future volatilities. The goals in this paper are: (i) to select an accurate forecasting procedure for predicting volatilities based on high frequency data from various standard models and modern prediction tools; (ii) to evaluate the predictive potential of those volatility forecasts for both the realized and the true latent volatility; and (iii) to quantify the differences using volatility forecasts based on high frequency data and using a GARCH model for low frequency (e.g. daily) data, and study its implication in risk management for two widely used risk measures. The pay-off using high frequency data for the true latent volatility is empirically found to be still present, but magnitudes smaller than suggested by simple analysis.

Customer lifetime value under complex contract structures

We analyze the problem of calculating the customer lifetime value (CLV) under contract structures that have an impact on customer dynamics. Typical examples are minimum contract durations, or fixed time points for contract cancellation. We show that classical Markov Chain models are not appropriate and may lead to large errors in the CLV. We propose a Semi-Markov formulation which leads to substantially better results.

Classification with gene expression data

A survey is given of tasks related to the construction and evaluation of classifiers applied to a renal cell cancer data set. Balanced sample splitting, non-specific filtering, linear discriminant analysis, nearest-neighbor prediction, and support vector machines are all concretely illustrated using the MLInterfaces package. Evaluations based on single and multiple random splits of data are compared. The entire presentation is given in a very generic programming format, to facilitate the adaptation and variation, by other investigators, of the techniques used here.