Seminar

Investigating the uncertainty caused by researchers' degrees of freedom in biostatistics, bioinformatics, and machine learning

M.Sc. Milena Wünsch, PhD student, Institut für Medizinische Informationsverarbeitung Biometrie und Epidemiologie (IBE), LMU,  Munich, Germany

Using flowR to improve the work with R code

M.Sc. Florian Sihler, Research Assistant, Institute of Software Engineering and Programming Languages, Ulm University, Germany

Two-sample hypothesis testing for high-dimensional and non-Euclidean data

Prof. Hao Chen, Associate Professor, Department of Statistics, University of California, Davis, USA

Bayesian compositional modelling of water quality in Scottish lochs

Prof. Mark J. Brewer, Director Biomathematics & Statistics Scotland (BioSS), The James Hutton Institute, Aberdeen, Scotland, UK

Probabilistic modelling of drug combination screens for precision oncology

Prof. Manuela Zucknick, Department of Biostatistics, Faculty of Medicine, University of Oslo, Norway

With high-throughput drug sensitivity screens we can quickly test compounds on cancer cell lines to determine treatment efficacy. Since molecular characterisation of the cell lines by various omics data sets is frequently available, we can link molecular features to treatment efficacy. The estimation of drug synergy is important when testing multiple compounds, but in vitro cell viability measurements can be imprecise due to measurement errors and limited number of data points, especially for drug combination experiments. We have earlier proposed a Bayesian model for drug response estimation with uncertainty quantification for an individual experiment, i.e. for one drug combination and one cell line. This model formed the basis for a probabilistic framework for dose–response prediction in high-throughput drug combination screens on large numbers of drugs and cell lines, which uses permutation invariant multi-output Gaussian processes. I will introduce the modelling framework and show how the model is able to learn from noisily observed measurements in settings where the underlying dose–response experiments are of varying quality, utilize different experimental designs, and the resulting training dataset is sparsely observed; and that it can accurately predict dose–response and capture relevant features indicating synergistic interaction between drugs.
Joint work with Leiv Rønneberg and Paul Kirk.

Additive Density-on-Scalar Regression

Prof. Sonja Greven, Chair of Statistics, School of Business and Economics, Humboldt-Universität zu Berlin, Germany

Samples of distributions are of interest in many applications, from income distributions in economics to distributions of soil contaminants in different states in geochemistry. Densities are useful summaries to work with, as they can be defined for continuous, discrete, mixed or even multivariate distributions and intuitively show bimodalities and areas of high and low probability mass. 
We present structured additive regression models to model densities given scalar covariates. To preserve nonnegativity and integration to one, we formulate our models for densities in so-called Bayes Hilbert spaces, enabling us to not only consider continuous densities, but also, e.g., discrete densities (compositional data) or mixed densities. Mixed densities occur in our application motivated by research on gender identity norms and the distribution of the woman's share in a couple's total labor income, as the woman's income share is a continuous variable having discrete point masses at zero and one for single-earner couples. We discuss interpretation of effect functions in our model via odds-ratios. We consider two data situations: First, where whole densities are observed and are directly used as responses. Second, when only samples or even individual realizations of the conditional distributions are observed, we use our additive regression approach to model the conditional density given covariates, nicely linking our models to distributional regression models. We derive consistency and asymptotic normality of the proposed (penalized) maximum likelihood estimator. To facilitate estimation, we show approximate equivalence of (penalized) likelihood and estimator of the original conditional density model to those of a certain Poisson additive model, greatly simplifying our implementation in an R package. We apply our framework to the motivating gender economic data from the German Socio-Economic Panel Study (SOEP) to model the distribution of the woman's share in a couple's total labor income, given year, place of residence and age of the youngest child. 

Genetic prediction modeling based on large cohort studies via statistical boosting

Hannah Klinkhammer, M. Sc., PhD Student at Institut für Medizinische Biometrie, Informatik und Epidemiologie (IMBIE), Universität Bonn
 

Optimizing Dose-Finding Studies for Drug Combinations Based on Exposure-Response Models

Andrew Hooker, Ph.D., Professor of Pharmacometrics, Department of Pharmacy; Pharmacometrics, Uppsala University, Sweden

(Joint) Modelling approaches for lung function decline and risk of death for cystic fibrosis patients

Prof. Dr. Elisabeth Bergherr, Chair of Spatial Data Science and Statistical Learning, Georg-August-Universität Göttingen

Multiple Endpoints and Prioritized Outcomes: Effects and Nonparametric Analysis Methods

Prof. Dr. Edgar Brunner, Emeritus, Department of Medical Statistics, Universitätsmedizin Göttingen

The analysis of multiple and composite endpoints in medical and biological trials is discussed since many years. Numerous procedures have been  developed for different situations and statistical models. Composite endpoints are  suggested to combine several outcomes into one single endpoint which can be evaluated by a simple analysis.
The danger, however, when using a combined endpoint is that the failure of an effect in an important clinical outcome may be masked by a large effect of a less important clinical outcome. Therefore, it has been suggested to prioritize the outcomes and if the outcome of a patient in the treatment group is better than that of a patient in the control group then score 1 is assigned or score 0 if it is worse. In case of no difference then one steps down to the next important outcome until no difference is found in the last important endpoint resulting in score ½. 
This handling of prioritized outcomes has been suggested by Buyse (2010) who called this procedure ‘generalized pairwise comparisons’ (GPC) since all pairs of patients from the treatment and the control group are compared by prioritizing the components. The resulting statistical procedure is closely related to the Mann-Whitney statistic.
To understand the advantages, drawbacks, and limitations of the GPC, the relations to rank procedures and U-statistics are presented and discussed. The practical use and the interpretation are demonstrated my means of examples.
Buyse M. Generalized pairwise comparisons of prioritized outcomes in the two-sample problem. StatMed. 2010; 3245- 3257.
 

Allelic expression imbalance to uncover disease-relevant gene regulation

Associate Professsor Michael I. Love, PhD, UNC School of Medicine Genetics, Chapel Hill, USA

Genome-wide association studies (GWAS) have identified tens of thousands of genomic loci that are associated with complex traits or diseases, many of which are located in non-coding regulatory regions. One potential mechanism by which allelic variation in these non-coding regions may affect phenotypes is that the variants may influence transcription in cis. A non-coding region that affects local transcription may be referred to as a cis-regulatory element (CRE). Individuals that are heterozygous at variants in CRE may exhibit imbalanced allelic expression at the regulated genes. With RNA-sequencing (RNA-seq) experiments, it is possible to observe such imbalance in allelic expression in the sequenced reads for those individuals also heterozygous for a variant in the exons of a regulated gene. In this talk I will give an overview of methods for detecting AEI from existing short-read RNA-seq data, and how new technologies will help in this task.

Sample size planning for multiple contrast test

Anna Pöhlmann, Institute for Biometrics and Clinical Epidemiology, Charité Berlin

Phases of development for statistical methods

Dr. Tim Morris, Principal Research Fellow MRC Clinical Trials Unit at UCL

Applied researchers need to be able to make informed decisions about which methods to use when. Research into statistical methods should thus provide an evidence-base that facilitates this. Unfortunately, the appetite of funders and journal editors for novelty and innovation in statistical methodology means the evidence base is often sparse. How should a method be developed and evaluated to go from ‘neat idea’ to trustworthy? We should be able to say when to use and avoid it, and how to make decisions for the study at hand.

By analogy to the familiar phases of drug-development, I will outline four ‘phases’ of research for statistical methods. This framework helps us consider the contribution of a piece of statistical research to our overall understanding. The aims of this view are to:
1. Help frustrated methodological researchers understand why their method has not been universally adopted;
2. Legitimise ‘late-phase’ statistical methods research;
3. Give applied researchers a way to articulate their hesitation about new methods.
I will give a few examples to illustrate, note some difficulties, and leave time for discussion.

Methods for the meta-analysis of ROC curves – A statistical challenge

Prof. Dr. Annika Hoyer, Head of Biostatistics and Medical Biometry Bielefeld University

Insights into the world of a CMC statistician

Dr. Beate Presser, Development Biologicals, Boehringer Ingelheim Pharma GmbH & Co. KG, Biberach


This presentation provides insights into the life of a CMC (Chemistry, Manufacturing, and Controls) statistician in the pharmaceutical industry, elucidating statistical questions and their solutions through various examples. These examples range from the concrete implementation of an ICH Q14 guideline on the development of analytical methods with DoE (design of experiment) to examples from the field of process modeling and unplanned data. Different model types are explained in practical examples.

Statistical Planning and Analysis of Translational Trials

Frank Konietschke, Acting Head of the Institute, Head of the Statistical Methods of Translation and Early Clinical Trials , Charité Berlin

How to better detect differentially expressed proteins when the sample size is small

Kai Kammers, Assistant Professor of Oncology, Johns Hopkins Medicine

Sparse group models for leveraging pleiotropy effects from GWAS

Benoit Liquet-Weiland, Professor of Mathematical and Computational Statistics, School of Mathematical and Physical Sciences, Macquarie University

Talk 1: Use of Environmental Mixture Methodology in Epidemiology Studies
Talk 2: Constructing DNA methylation risk scores for mediation analysis

Grace Christensen and Junyu Chen, MPH PhD Candidate Emory University’s Rollins School of Public Health

Talk 1: EFSA’s guidance on the use of the BMD approach in risk assessment
Talk 2: The statistical methodology behind EFSA’s BMD platform
Talk 3: Illustration of EFSA’s new BMD analysis platform

Prof. Dr. Marc Aerts, Hasselt University Belgium,  and Dr. José Cortiñas Abrahantes, EFSA European Food Safety Authority

Efficient Designs for toxicological experiments

Weng Kee Wong, Professor at the Department of Biostatistics, University of California, Los Angeles

I first discuss the importance and challenges in toxicology problems, which are increasingly pervasive in our daily lives.  A main focus of the talk is how to implement more informative designs in the laboratory that will likely lead to more accurate statistical inference.  Optimal experimental design theory and ideas will be used to construct model-based optimal designs for a variety of common nonlinear models used in toxicology and I will present efficient designs to estimate model parameters, or, one or more functions of the model parameters, including optimal designs to identify thresholds, detect presence of hormesis, and in identifying an adequate model among several postulated models.  In addition, I will discuss metaheuristics as general optimization tools to find optimal designs for more complex problems and demonstrate how to them in web-based tools to find optimal designs.  If time permits, I will also discuss one of my current work in toxicology that concerns Haber’s law.
 

How does air pollution effect our brain? The complex relationship between environmental, social and epigenetic factors?

Anke Huels, Assistant Professor, Department of Epidemiology and Gangarosa Departmentof Environmental Health, Rollins School of Public Health, Emory University, Atlanta, USA

Prediction Intervals for overdispersed Poisson data

Dr. Max Menssen, Institute of Cell Biology and Biophysics, Leibniz Universität Hannover

Novel clinical trial designs for precision oncology:

Prof. Dr. Annette Kopp-Schneider,    Deutsches Krebsforschungszentrum in der Helmholtz-Gemeinschaft, Heidelberg

The aim of precision oncology is to provide a targeted therapy for each individual tumor. To reach this goal, it is necessary to identify in each patient actionable biomarkers and to determine a matched targeted therapy for the specific biomarker(s). Clinical trials are necessary to investigate whether the biomarker-treatment combination in fact has the desired effect. Due to the large number of possible biomarker-treatment combinations, and due to the small patient numbers matching such a combination, classical clinical phase III trials are not feasible in this situation. Precision oncology therefore requires the development of novel trials designs. We will discuss the concepts of basket, umbrella and platform designs for precision oncology and will present the challenges for the statistical analysis in these trial designs. Bayesian methods are naturally suited for precision oncology clinical trials as they allow for more flexible designs and they are suited for incorporation of external information and for borrowing of information between arms of clinical trials. We will discuss the use of Bayesian methods for clinical trials and present some current developments.
 

Optimal test procedures for multiple hypotheses controlling the familywise expected loss

Prof. Dr. Frank Bretz, Novartis Pharma AG, Biostatistics & Statistical Reporting, CH-4002 Basel, Switzerland

Genotoxicity Assessments of Pharmaceuticals

Prof. Dr. Helena Geys, Hasselt University, Belgium

A novel group-sequential phase II design for clinical trials with binary endpoints based on Bayesian evidence values

Dr. Riko Kelter, Department of Mathematics, Research Group Stochastics, University of Siegen

A replication crisis in methodological statistical research?

Prof. Dr. Anne-Laure Boulesteix, Institut für Medizinische Informationsverarbeitung Biometrie und Epidemiologie (IBE), Fakultät der LMU München

Shinybrms and projpred: Introductory and advanced Bayesian (and not-so-Bayesian) statistics in practice

Frank Weber, Institut für Biostatistik und Informatik in Medizin und Alternsforschung (IBIMA)

Virtual Control Groups: Using historical toxicity data to replace control-group animals

Alexander Gurjanov, PHD Student at Bayer Pharmaceuticals

Talk 1: An Intro to ToxicR and Bayesian Dose-Response
Talk 2: Fast Bayesian Gaussian Process Regression Using Compression H-Matrices

Matt Wheeler, Ph.D. National Institute of Environmental Health Sciences
 

The Hazelwood Health study and other opportunities in Australia

Prof. Michael Abramson, School of Public Health & Preventive Medicine, Monash University, Australia

Random forests on high-dimensional data: from classification and survival analysis to generative modelling

Prof. Dr. Marvin N. Wright, Leibniz-Institut für Präventionsforschung und Epidemiologie – BIPS, Bremen

Issues in supervised prediction and classification for complex chemical substances

Prof. Fred Wright, Ph.D., Departments of Statistics and Biological Sciences, North Carolina State University, USA

Modeling Heterogeneity for Stratified Populations

Prof. Menggang Yu, Department of Biostatistics and Medical informatics, University of Wisconsin-Madison, USA

Propensity Score: an Alternative Method of Analyzing Intervention Effect

Prof. Dr. sc. hum. Oliver Kuß, Deutsches Diabetes-Zentrum, Düsseldorf

Gene expression to Omics: The evolution of genetic research

Ashtyn Areal, IUF Düsseldorf

Analysis and design of clinical trials with biologics using dose-time-response models

Markus Lange, Novartis

The case time series design for high-dimensional data analyses

Prof. Antonio Gasparrini, London School of Hygiene and Tropical Medicine, London, UK

Mehrstadien-Modelle in der Epidemiologie chronischen Erkrankungen

Prof. Dr. rer. nat. Ralph Brinks , Lehrstuhl für Medizinische Biometrie und Epidemiologie, Private Universität Witten/Herdecke gGmbH

Assessing interactive effects of air pollution and temperature– approaches and challenges?

Dr. Susanne Breitner , Helmholtz-Zentrum München

Dose- response analysis in toxicology: considering dose both as qualitative factor and quantitative covariate (using R)

Prof. Dr. Ludwig Hothorn, Biostatistician, Retired from Leibniz University Hannover

Systems genetics strategies for describing the transcriptional connectome and it's role in complex traits

Associate Professor Laura Saba, Skaggs School of Pharmacy and Pharmaceutical Sciences, University of Colorado, USA

Can statistics save preclinical research?

Prof. Dr. med. Ulrich Dirnagl, Abteilungsdirektor Experimentelle Neurologie, Charité Berlin

Dose-response analysis for gene-expression data

Scott S. Auerbach, PhD, und Matt Wheeler, PHD, Institute of Environmental Health Sciences, Durham, NC, USA

Abstract

Good Scientific Practise for Doctoral Researchers

Dr. Peter Schröder, brain4hire, Graduiertenzentrum TU Dortmund

Recent extensions on boosting for statistical modelling

Prof. Dr. Andreas Mayr, Head of WG Statistical Methods in Epidemiology, Universität Bonn

Polygenic risk scores – Applicability beyond risk prediction and for different omics data

Anke Hüls, PhD, MSc, Assistant Professor, Department of Epidemiology and Gangarosa Department of Environmental Health, Emory University

Advances in dose-response analysis

Prof. Christian Ritz , National Institute of Public Health,Copenhagen