The UNIVERSITY OF EDINBURGH SCHOOL OF MATHEMATICS --------------------------------------------------------------- Date: Friday 23 November 2012 at 3pm Location: 5326 JCMB ---------------------------------------------------------------- Silvia Liverani (Imperial College London) "A new Bayesian Dependence Model for Regulatory Networks" Abstract The usual assumption for clustering is that observations in the same cluster share the same expression profile - accounting for measurement error - and are expressed independently of observations outside their own cluster. However, when searching over certain types of data, it is unrealistic to assume independence between clusters as, for instance, in the context of microarray experiments, it is well known that genes belong to regulatory pathways and activate and inhibit each other. In this talk we present a formal methodology for developing the semantics of new Gaussian graphical models so that these are customised to be faithful to a given scientific domain. This new class of models is designed to utilize as much as possible the convenient separation properties of a Bayes cluster model whilst continuing to respect the types of dependence that scientists might conjecture exist. We demonstrate how this class of models can be generalized to encompass more credible deterministic functional dependence structures associated with regulation. We show that it is possible to search over a large number of different hypothesized regulation models, just as in Bayes cluster models, and we interpret the results in a causal framework that provides the scientist with a new hypothesis-generating tool. We demonstrate the efficacy of these methods on a class of circadian models where interest focuses on the regulatory relationships between different genes. Each member in the class can be evocatively labelled by a regulation graph closely resembling in its semantics the graphical representations of regulation familiar to biologists. This is joint work with Jim Q. Smith (University of Warwick). There will be tea and coffee after the talk. The seminar website (in google calendar): http://www.maths.ed.ac.uk/events/statistics This seminar series is a part of Maxwell Institute seminar series. -- 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.