BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Selection of the Regularization Parameter in Graphical Models usin g Network Characteristics - Natalia Bochkina (University of Edinburgh) DTSTART:20160727T130000Z DTEND:20160727T133000Z UID:TALK66860@talks.cam.ac.uk CONTACT:INI IT DESCRIPTION:We study gene interaction networks using Gaussian graphical mo dels that represent the underlying graph structure of conditional dependen ce between random variables determined by their partial correlation or pre cision matrix. In a high dimensional setting\, the precision matrix is est imated using penalized likelihood by adding a penalization term which cont rols the amount of sparsity in the precision matrix and totally characteri zes the complexity and structure of the graph. The most commonly used pena lization term is the L1 norm of the precision matrix scaled by the regular ization parameter which determines the tradeoff between sparsity of the gr aph and fit to the data. We propose several procedures to select the regul arization parameter in the estimation of graphical models that focus on re covering reliably the appropriate network characteristic of the graph\, an d discuss their Bayesian interpretation. Performance is illustrated on sim ulated data as well as on colon tumor gene expression data. This work is e xtended to reconstructing a differential network between two paired groups of samples.  \;  \;

This is joint work with with Adria Caballe Mestres (University of Edinburgh and Biomathematics and Stat istics Scotland) and Claus Mayer (Biomathematics and Statistics Scotland). LOCATION:Seminar Room 1\, Newton Institute END:VEVENT END:VCALENDAR