| 14.00 - 14.05 | Velkomst |
| 14.05 - 14.50 |
Torben Martinussen, Institut for Sundhedstjenesteforskning, Odense: The Aalen Additive Frailty Hazards Model Copies of slides The Cox model is used very often when analyzing survival data. It requires constant relative risks, which may be violated in practice, however. A very useful alternative is the Aalen additive hazards model that is very flexible allowing for time-varying covariate effects. In this talk we consider correlated survival data and how to estimate in the Aalen additive frailty hazards model. This problem has to our knowledge not been considered before. Large sample results are provided and shown to work well in small samples. We conclude with a univariate data example where converging hazards is an issue indicating that a frailty model might be an appealing model to use. One of the time-varying covariate effects does not seem to be well explained by the usual Cox frailty model approach so it is tempting to try to use the Aalen additive frailty hazards model in this application as well. |
| 14.50 - 15.20 | Kaffe og kage (bygn. 1150) |
| 15.20 - 16.10 |
David Edwards, Institut for Genetik og Bioteknologi
,Foulum: Sequentially rejective test procedures for partially ordered sets of hypotheses Copies of slides Clinical trials often address multiple hypotheses, for example relating to multiple treatments and/or endpoints. Two widely used approaches to control multiplicity in such trials are closed test procedures and fixed sequence test procedures. In the former the hypotheses are taken to be unordered and in the latter the hypotheses are completely ordered. This talk describes how these approaches may be combined and generalized to deal with partial orderings of the set of hypotheses. This enables inference strategies to be constructed that more closely mirror the study objectives. Various examples are given. The talk is based on joint work with Jesper Madsen, namely Edwards D, Madsen J (2007) Constructing multiple test procedures for partially ordered hypothesis sets, Statistics in Medicine. |
| 16.10 - 17.00 |
Anders Tolver Jensen, Institut for Grundvidenskab og Miljø/Statistik, KU: Drug Discovery Based on Combinatorial Chemical Libraries Copies of slides Screening large chemical libraries plays an important role in the preliminary phase of drug discovery. Newly developped techniques based on combinatorial chemistry and principles from biological evolution allow to set up experiments the address the ability of more than one billion drugs to bind to a specific target molecule in one single experiment. But how should we approach the multiple testing problem of selecting candidate drugs that bind to a particular target? A good mathematical model for the individual steps of the chemical process that takes into acount the combinatorial structure of the chemical libraries is crucial to exploit the data in a reasonable way. This may provide us with useful insight that allows us to discuss how to design future experiments and optimize the various parameters of the chemical processes in order to increase the power of the multiple testing procedure. This is joint work with Ib Michael Skovgaard and an external company. |
| 17.15 - 18.00 | Øl og sodavand i vandrehallen Matematisk Institut bygn. 1530 |
| 18.00 - 24.00 | Middag i Matematisk kantine bygn. 1536 |
| 9.00 - 9.45 |
Jakob Grove, Afdeling for Epidemiologi, Århus: A Pedestrian Look at Genetic Epidemiology Copies of slides Based on consideration we have had in connection with a number of case-control based candidate gene studies we are conducting, I will walk through some of the basic concepts of genetics and how they can be analyzed statistically. In the process we will touch upon some of the problems in this area of statistics. Being candidate gene studies we are dealing with 100s rather than 1000s of SNPs which put us in the framework of classical statistical genetics, but it still leaves us with substantial logistical challenges. - People new to statistical genetics/genetic epidemiology can start here. |
| 9.45 - 10.30 |
Asger Hobolth, BioInformatics Research Center, Århus: Statistical analysis of DNA sequence evolution Copies of slides Continuous-time Markov chains provide a stochastic description of how DNA sequences evolve over time. The complexity of the description varies from the simple site-independent nucleotide substitution models to the very challenging context-dependent models. Both types of models can be analysed using the EM-algorithm. However, for the context-dependent model the E-step is no longer analytically tractable, but must be performed using simulations. An efficient solution to this simulation problem is discussed in detail. The talk is based on joint work with Jens Ledet Jensen and Søren Asmussen (both Aarhus University), and Eric Stone and Jeff Thorne (both North Carolina State University). |
| 10.30 - 11.00 | Kaffe og kage (bygn. 1150) |
| 11.00 - 11.45 |
Simon Myers, Department of Statistics, Oxford: Sequence motifs, recombination hotspots, and genome evolution in humans Copies of slides |
| 11.45 - 12.30 |
Luc Janss, Institut for Genetik og Bioteknologi
,Foulum:
Constructing prediction models from very many predictors - a (somewhat) new philosophy - Copies of slides The latest advancement in DNA genotyping methodology is the introduction of SNP (Single Nucleotide Polymorphism) arrays, which allow to determine 20-50K marker genotypes on an individual at affordable costs. The statistical analysis of these SNP arrays falls in the category of “large p small n problems”, which are so far tackled with mixed models and two main versions of Bayesian models. One application of these SNP arrays is the construction of “Genomic Predictions” which attempt to predict the total genetic level of an individual. Research has indicated that a model selection philosophy without formal selection of SNP markers usually works best to make these predictions: a Bayesian Variable Selection Method which only makes a “fuzzy” selection, or a mixed linear or Bayesian hierarchical model in which all SNP markers are used simultaneously without any selection. The use of these Genomic Predictions is now expected to drastically change genetic improvement programs for agricultural animal and plant populations. The “no selection” philosophy could possibly also be applied to improve predictors from microarray gene expression data and to improve predictors for total genetic risk for diseases in humans. |
| 12.30 | Sandwich og afgang |