Next statistics seminar will take place on
22 November at 3:10pm, room JCMB5323.
Michael Papathomas, University of St Andrews
Title: Exploring the presence of complex dependence structures in
epidemiological and genetic data
Abstract:
Detecting high-order interactions between covariates in a linear model
framework is not straightforward, due to the difficulty in investigating
an unwieldy large space of competing models. First, we present profile
regression, which reduces dimensionality by using as the basic unit of
inference the subject’s covariate profile. The profiles are clustered
into groups using the Dirichlet process, and are associated via a
regression model to a relevant outcome. Post-processing and model
averaging techniques are used to create interpretable output. Variable
selection is incorporated in the modelling, to allow for the exploration
of high-dimensional data from association studies. Secondly, we examine
the relation between the presence of interaction terms in linear
graphical models, and variable selection within clustering. We utilize a
theoretical result on the relation between clustering and log-linear
modelling to detect factors that do not form interaction terms. Removing
these factors from the analysis leads to the reduction of a covariate
space that would otherwise translate to a sparse contingency table, thus
making the implementation of standard model search algorithms like the
Reversible Jump feasible.
There will be wine after the talk.
This seminar is joint with BioSS, and it is a part of Maxwell institute
seminar series.
Website: http://www.maths.ed.ac.uk/events/statistics
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|