- statslausanne - lists.sib.swiss

statistical forum October 30
by Rousson Valentin 22 Oct '14

22 Oct '14

Dear all, The next seminar in our series of statistics talks for people interested in quantitative methods in biology and medicine will take place on Thursday 30 October 2014 at 11h00 in the Salle Delachaux (Biopôle 2, premier étage, Route de la Corniche 10, Lausanne, M2: Vennes). The speaker will be Dr. Aaron McDaid (IUMSP, postdoc by Zoltan Kutalik) who will speak about MCMC for mixtures of Gaussians and model selection Abstract: In cluster analysis, it is usual to consider mixtures of Gaussian distributions. In this presentation, we consider six multivariate models of such mixtures which differ in the restrictions placed on the covariance matrices. The covariance may be different in each cluster or constrained to be equal. The covariance may be restricted to be diagonal-only or, further, such that each element of the diagonal is equal (spherical clusters). The most general covariance model places no restriction on the covariance matrices. Estimating the number of clusters and selecting one from the six covariance models in a fully Bayesian framework is the main goal of this research. Through the use of integration and conjugate priors, such as the Wishart distribution for covariances, we are able to compute the joint probability of the number of clusters, the assignment vector and the observed data, given a particular choice from the six covariance models, where all other quantities (means, covariances, mixing proportions) have been integrated out. We have used that joint probability in the `allocation sampler', an existing MCMC algorithm where we have included some novel moves to speed it up, to sample estimates for the number of clusters and for the assignment vector. This allowed a direct estimation of the number of clusters, freeing us from the need to run the algorithm repeatedly, once for each possible number of clusters. The challenging problem was then to select one model from the six covariance models. We show how we can run the six MCMC chains in parallel and combine samples from the six chains, in proportion to the Bayes Factor, using an augmented data model based on a method by Carlin and Chib. This algorithm is illustrated and compared to other classical approaches of cluster analysis using six real data sets. Hope to see you there! Valentin (for the organizers: Jérôme Goudet, Valentin Rousson, Frédéric Schütz)

1 0

Modelling Cancer Screening Interventions (Thursday 19 June, 10:00)
by Rousson Valentin 10 Jun '14

10 Jun '14

Dear all, At the occasion of the symposium « Dépistage mammographique organisé : état des lieux et perspectives en Suisse » organized by the "Ecole Romande de Santé Publique", in collaboration with the Institute for Social and Preventive Medicine of Lausanne and the Institute for Global Health of Geneva, Professor Harry J de Koning (Erasmus Medical Center, Rotterdam, The Netherlands) will present a conference entitled "Modelling Cancer Screening Interventions", which will take place on Thursday 19 June 2014 at 10:00 in the salle Delachaux (Biopôle 2, first floor, route de la Corniche 10, 1010 Lausanne, M2 station: Vennes; note the unusual time 10:00). Abstract: Population-based cancer screening is a major effort to try and reduce mortality and morbidity of the disease. The European code against cancer at present stipulates screening for 3 cancer sites: cervical cancer age 30 years and over by a Pap smear, breast cancer age 50 and over by mammography and colorectal cancer screening from age 50 and over with faecal occult blood test. Many countries’ public health policies endorse these guidelines, but differences exist in interval, age range and tests applied. Randomised controlled trials or rather clear-cut geographical differences from the very past have provided evidence on the benefits of cancer screening, but these often give answers on one specific design. Modelling is often needed to estimate the long-term impact on benefits, harms, cost and savings, and to guide policies. At present, debate exists whether new treatment modalities may be so effective, that benefits seen in the population as a whole, may be contributed to treatment rather than screening (and early treatment). Sometimes, screening is considered expensive, given the annual number of screens needed. However, new screening techniques may lead to alterations and more effective screening programmes. It may also lead to new guidelines, e.g., regarding prostate and lung cancer screening, or screening in high risk groups. In this lecture, modelling of the Canadian breast cancer screening trial will be presented, to estimate (evidence on) effectiveness, and modelling lung cancer screening to guide policy. Hope to see you there! Jean-Luc Bulliard et Valentin Rousson ---------------------------------------------------- Unité de Statistique Institut Universitaire de Médecine Sociale et Préventive Centre Hospitalier Universitaire et Université de Lausanne Biopôle 2 Route de la Corniche 10 CH-1010 Lausanne Tél : ++41 (021) 314 73 28 Fax : ++41 (021) 314 73 73 Email: Valentin.Rousson(a)chuv.ch<mailto:Valentin.Rousson@chuv.ch>

1 0

Statistical forum on June 5
by Rousson Valentin 28 May '14

28 May '14

Dear all, The next seminar in our series of statistics talks for people interested in quantitative methods in biology and medicine will take place on Thursday 5 June 2014 at 11h00 in the Salle Delachaux (Biopôle 2, premier étage, Route de la Corniche 10, Lausanne, M2: Vennes). The speaker will be Athanassios Kondylis (Philippe Morris International, Neuchâtel) who will speak about "An overview on PLS methods in regression & some recent advances" Abstract: PLS methods are very suitable methods in regression and classification when dealing with collinear and high dimensional data (n<p). Such data sets are very common in modern statistical practice. An overview of PLS methods will be provided, with focus given on regression problems. Algorithmic, numerical and statistical aspects of PLS regression will be presented. Within the framework of regularized regression, PLS regression will be compared to principal components and to ridge regression. Some recent advances in PLS regression highlighting its relation to principal components and simplifying the interpretation of the PLS solution will be finally given. Data sets coming from chemometrics and chemoinformatics will be used to illustrate the presented methods. Hope to see you there! Valentin (for the organizers: Jérôme Goudet, Valentin Rousson, Frédéric Schütz) ---------------------------------------------------- Valentin Rousson Unité de Statistique Institut Universitaire de Médecine Sociale et Préventive Centre Hospitalier Universitaire et Université de Lausanne Biopôle 2 Route de la Corniche 10 CH-1010 Lausanne Tél : ++41 (021) 314 73 28 Fax : ++41 (021) 314 73 73 Email: Valentin.Rousson(a)chuv.ch<mailto:Valentin.Rousson@chuv.ch>

1 0

Statistical forum on May 1
by Rousson Valentin 23 Apr '14

23 Apr '14

Dear all, The next seminar in our series of statistics talks for people interested in quantitative methods in biology and medicine will take place on Thursday 1 Mai 2014 at 11h00 in the Salle Delachaux (Biopôle 2, premier étage, Route de la Corniche 10, Lausanne, M2: Vennes). The speaker will be Daniel Wegmannn (University of Fribourg, Switzerland) who will speak about Model based inference from genomic data - Approximate and some less approximate attempts Abstract: The rapidly increasing amount of available sequencing and genotyping data of huge samples triggers the desire to contrast or infer parameters from ever more complex models. Unfortunately, the likelihood functions of such models are often unknown, intractable or simply computationally out of reach. Here I will present two major strategies in such cases through examples: approximate inference and full likelihood inference of approximate models. I will begin by presenting the Approximate Bayesian Computation framework and show an application to GWAS data. I will then present two ongoing research projects where we use full likelihood inference of approximate models, firstly a novel approximate Wright-Fisher process to infer parameters from genetic data obtained at different time points, and secondly an attempt to formulate a model to infer gene-trait associations along with population structure while taking linkage partly into account. Hope to see you there! Valentin (for the organizers: Jérôme Goudet, Valentin Rousson, Frédéric Schütz) ---------------------------------------------------- Valentin Rousson Unité de Statistique Institut Universitaire de Médecine Sociale et Préventive Centre Hospitalier Universitaire et Université de Lausanne Biopôle 2 Route de la Corniche 10 CH-1010 Lausanne Tél : ++41 (021) 314 73 28 Fax : ++41 (021) 314 73 73 Email: Valentin.Rousson(a)chuv.ch<mailto:Valentin.Rousson@chuv.ch>

1 0

statistical forum on April 3
by Rousson Valentin 28 Mar '14

28 Mar '14

Dear all, The next seminar in our series of statistics talks for people interested in quantitative methods in biology and medicine will take place on Thursday 3 April 2014 at 11h00 in the Salle Delachaux (Biopôle 2, premier étage, Route de la Corniche 10, Lausanne, M2: Vennes). The speaker will be Georg B. Ehret (University of Geneva and Center for Complex Disease Genomics, Johns Hopkins University, USA) who will speak about Blood pressure genomics - methods to improve the discovery of blood pressure variants Abstract: Genome-wide association studies (GWAS) have contributed new knowledge on blood pressure (BP) elevation, a major cardiovascular risk factor: Twenty-nine BP SNPs identified in up to 203,056 individuals (Levy Nature Gen 2009, Newton-Cheh Nature Gen 2009, Ehret Nature 2011) have shown a clinically significant BP effect size, are associated with BP in multiple ethnicities, and predict target organ damage, excepting the kidney. The International Consortium for Blood Pressure Genome-Wide Association Studies and parent- & related consortia have undertaken a number of new analyses, including studies using refined phenotypes, interaction analyses, and studies with larger sample sizes that improve the discovery of BP variants. Hope to see you there! Valentin (for the organizers: Jérôme Goudet, Valentin Rousson, Frédéric Schütz)

1 0

statistical forum on March 20
by Rousson Valentin 13 Mar '14

13 Mar '14

Dear all, The next seminar in our series of statistics talks for people interested in quantitative methods in biology and medicine will take place on Thursday 20 March 2014 at 11h00 in the Salle Delachaux (Biopôle 2, premier étage, Route de la Corniche 10, Lausanne, M2: Vennes). The speaker will be Marie-Annick Le Pogam (IUMSP) who will speak about Statistical concerns about quality and safety indicators based on hospital inpatient administrative data Abstract: Reflective of an increased demand for transparency and accountability, it has become common practice world while to release quality report-cards comparing hospital-specific outcomes. Health authorities are interested in identifying any providers whose outcome deviates substantially from some standard or target (which can be the national average), after accounting for confounding factors beyond their control, primarily patient case-mix. Furthermore, in order to control health expenditures in the context of both the global financial and economic crisis and the increase of health spending due to global ageing, some governments tried to improve hospital efficiency using pay-for-performance incentive programs that link quality to payment. Within this context, providing valid and reliable outcome measures of hospital care at a national level has become challenging for researchers. Because older patients are subject to frequent and numerous comorbidities, polypharmacy, impaired hepatic and renal function and both functional and cognitive decline, their admission to acute care hospitals are at high risk or adverse events (AE). As in many countries, the Swiss Federal council has endorsed the WHO Health 2020 policy framework for health and well-being in order to align Swiss health system optimally with future challenges. The main objectives that shape the action plan include 1) the production of quality and safety indicators – focusing on outcome indicators - to inform health authorities’ decisions, 2) the public reporting of healthcare information to inform patients’ choice, 3) health services’ benchmarking to improve efficiency and quality of care. The AHRQ patient safety indicators (PSIs) are now broadly used to estimate the occurrence of potential adverse events (AE) and in-hospital complications following surgeries, medical procedures and childbirth. PSIs provide hospital AE or complication adjusted-rates at a national level using routinely collected hospital discharge abstract administrative data. In my PhD project, we hypothesize that AHRQ PSIs based on Swiss inpatient administrative data could provide valid and reliable screening and assessment tools for elderly patient safety in acute care hospitals. They should be “refined” in order to consistently identify potentially “outlying” hospitals according to their complication rates and therefore of interest for further investigation. Empirical statistical analysis will have to face several methodological issues that are still discussed in the international scientific literature: • The choice between fixed and random effects models and between frequentist and Bayesian approach to improve the validity and reliability of safety measures for hospital profiling; • The effect of unmeasured differences in case-mix (due to the quality of administrative data and data availability) on their validity and reliability; • The interpretation of unexplained variability between hospitals after adjusting for some hospital characteristics (and for patient case-mix); • The ways of dealing with both “overdispersion” and “excess zeros” of hospital raw data or rates; • And methods to account simultaneously for various hospital sizes (and more especially for small hospitals). Hope to see you there! Valentin (for the organizers: Jérôme Goudet, Valentin Rousson, Frédéric Schütz) ---------------------------------------------------- Unité de Statistique Institut Universitaire de Médecine Sociale et Préventive Centre Hospitalier Universitaire et Université de Lausanne Biopôle 2 Route de la Corniche 10 CH-1010 Lausanne Tél : ++41 (021) 314 73 28 Fax : ++41 (021) 314 73 73 Email: Valentin.Rousson(a)chuv.ch<mailto:Valentin.Rousson@chuv.ch>

1 0

(Another) Seminar by Terry Speed on Fri 7 March 2014 (15:00) : "Removing unwanted variation: from principal components to random effects ?"
by Frédéric Schütz 19 Feb '14

19 Feb '14

Following yesterday's email, Terry Speed will give a second seminar on 7 March afternoon, at EPFL. See the full description below. *EPFL Statistics Seminar* Friday, 7 March 2014 - 15h00 (sharp). CE2 (EPFL) *Removing unwanted variation: from principal components to random effects.* Professor Terry Speed UC Berkeley & Walter and Eliza Hall Institute (Hosts: Anthony Davison, Darlene Goldstein, Victor Panaretos) Abstract: *Ordinary least-squares is a venerable tool for the analysis of scientific data originating in the work of A-M. Legendre and C. F. Gauss around 1800. Gauss used the method extensively in astronomy and geodesy. Generalized least squares is more recent, originating with A. C. Aitken in 1934, though weighted least squares was widely used long before that. At around the same time (1933) H. Hotelling introduced principal components analysis to psychology. Its modern form is the singular value decomposition. In 1907, motivated by social science, G. U. Yule presented a new notation and derived some identities for linear regression and correlation. Random effects models date back to astronomical work in the mid-19th century, but it was through the work of C. R. Henderson and others in animal science in the 1950s that their connexion with generalized least squares was firmly made. These are the diverse origins of our story, which concerns the removal of unwanted variation in high-dimensional genomic and other "omic" data using negative controls. We start with a linear model that Gauss would recognize, with ordinary least squares in mind, but we add unobserved terms to deal with unwanted variation. A singular value decomposition, one of Yule's identities, and negative control measurements (here genes) permit the identification of our model. In a surprising twist, our initial solution turns out to be equivalent to a form of generalized least squares. This is the starting point for much of our recent work. In this talk I will try to explain how a rather eclectic mix of familiar statistical ideas can combine with equally familiar notions from biology (negative and positive controls) to give a useful new set of tools for omic data analysis. Other statisticians have come close to the same endpoint from a different perspectives, including Bayesian, sparse linear and random effects models. Joint work with Johann Gagnon-Bartsch and Laurent Jacob.*

1 0

SIB Workshop on "Reproducible Research", April 4 2014
by Frédéric Schütz 19 Feb '14

19 Feb '14

Dear all, the SIB Swiss Institute of Bioinformatics is organizing a new 1-day workshop on "Reproducible Research" which will be held in Lausanne on the *4 April 2014*. It is likely to be of interest to several readers of this mailing-list -- "reproducible research" was the topic of our first meeting two years ago, and at the time, several participants indicated their interest in the organization of such a workshop. Full description: This course will focus on the concept and techniques that allow statisticians/data analysts to ensure as much as possible that their analysis performed using R is reproducible. Reproducible research in statistical analysis is the idea that it should be possible for other people to reproduce a published data analysis and obtain the same results ("other people" also include the person who did the initial data analysis, 6 months later). While all data analysts agree that the concept of reproducible research is important, very few of them actually take active steps to ensure that their work is reproducible. Participants should have experience in using R for analyzing data. Basic knowledge of LaTeX or LyX is an advantage. Participants should bring a laptop and be able to install custom applications. A list of required applications including R, LaTeX and GIT will be provided before the course. The course is organized by the Bioinformatics Core Facility (BCF) of the SIB Swiss Institute of Bioinformatics. Further information and registration are available from http://edu.isb-sib.ch/course/view.php?id=155 Please note that the number of seats is limited. Hope to see you there! Frédéric -- Frédéric Schütz, PhD Frederic.Schutz(a)isb-sib.ch Maître d'Enseignement et de Recherche, Center for Integrative Genomics Senior Statistician, Bioinformatics Core Facility (Delorenzi group) Swiss Institute of Bioinformatics http://www.isb-sib.ch/ University of Lausanne http://www.unil.ch/ Batiment Genopode Quartier Sorge Phone: +41 21 692 40 94 CH-1015 Lausanne Fax: +41 21 692 40 65

1 0

Seminar by Terry Speed on Fri 7 March 2014: "Normalization of RNA-Seq Data: Are the ERCC Spike-In Controls Reliable ?"
by Frédéric Schütz 18 Feb '14

18 Feb '14

Terry Speed from the Walter and Eliza Hall Institute, Melbourne, AUS and the University of California, Berkeley, USA, will talk about: "Normalization of RNA-Seq Data: Are the ERCC Spike-In Controls Reliable?" Friday, March 7th, 2014 at 11:00 Génopode Auditorium B (UNIL, Dorigny campus) Abstract: Joint work with Sandrine Dudoit, Davide Risso and John Ngai, all from UC Berkeley. The External RNA Control Consortium (ERCC) developed a set of 92 synthetic polyadenylated RNA standards that mimic natural eukaryotic mRNA (Jiang et al., 2011). The standards are designed to have a wide range of lengths (250-2,000 nucleotides) and GC-contents (5-51%). The ERCC standards can be spiked into RNA at various concentrations prior to the library preparation step and serve as negative and positive controls in RNA-Seq. Ambion commercializes spike-in control mixes, ERCC ExFold RNA Spike-in Control Mix 1 and 2, each containing the same set of 92 standards, but at different concentrations. We investigate the use of the ERCC spike-in controls for two main purposes: (a) Quality assessment/quality control (QA/QC) of RNA-Seq data and benchmarking of normalization and differential expression (DE) methods, and (b) Direct inclusion in between-sample normalization procedures. We have two RNA-seq data sets which make use of the ERCC controls: a local one concerning treated and untreated zebrafish tissue, and some of the SEQC samples. A variety of normalization methods will be compared, both using and not using the ERCC controls.

1 0

Statistical Forum November 28 2013, 11:00
by Rousson Valentin 19 Nov '13

19 Nov '13

Dear all, The next seminar in our series of statistics talks for people interested in quantitative methods in biology and medicine will take place on *Thursday 28 November 2013 at 11h00* in the Salle Delachaux (Biopôle 2, premier étage, Route de la Corniche 10, Lausanne, M2: Vennes). The speaker will be *Alina Matei* (University of Neuchâtel) who will speak about "Unequal probability sampling designs, entropy and sample coordination in finite populations" The slides will be in English and the talk in French. Abstract: After a short introduction to unequal probability sampling designs, we show how the sample randomness is related to the Shannon entropy. Higher entropy of a sampling design implies more randomness in sample selection. We present afterwards the unequal probability sampling design, which maximizes the entropy criterion in the class of all possible sampling designs having the same unit inclusion probabilities and a fixed sample size (called maximum fixed-size entropy sampling or Conditional Poisson sampling). Next, we focus on sample coordination, which seeks to create a probabilistic dependence between two or more samples drawn successively or simultaneously from overlapping finite populations. This dependence leads to a maximization or a minimization of the sample overlap. The former is known as positive coordination, and the latter as negative coordination. Positive coordination is mainly employed for estimation purposes and for reasons related to data collection costs. Negative coordination is mainly used to diminish the response burden of the sampled units selected in different samples. We show two schemes to coordinate Conditional Poisson samples. The proposed methods are evaluated using the size of the expected sample overlap, and compared with their competitors, by means of the Monte Carlo simulation. Hope to see you there! Valentin (for the organizers: Jérôme Goudet, Valentin Rousson, Frédéric Schütz) ---------------------------------------------------- Valentin Rousson Unité de Statistique Institut Universitaire de Médecine Sociale et Préventive Centre Hospitalier Universitaire et Université de Lausanne Biopôle 2 Route de la Corniche 10 CH-1010 Lausanne Tél : ++41 (021) 314 73 28 Fax : ++41 (021) 314 73 73 Email: Valentin.Rousson(a)chuv.ch<mailto:Valentin.Rousson@chuv.ch>

1 0