Tid: 9 december 2002 kl 1515-1700
Plats : Seminarierummet 3733, Institutionen för matematik, KTH, Lindstedts väg 25, plan 7. Karta!
Föredragshållare: Ola Hössjer, Matematisk statistisk, Stockholms Universitet.
Titel: Computing genomwise significance levels in linkage analysis.
Sammanfattning: Statistical linkage analysis is a method for localizing genes causing or contributing to an inheritable disease. The idea is to test, at each chromosomal position, whether inheritance of the disease is correlated (linked) to inheritance of DNA at that position. Because of the large number of tests being performed, it is important to adjust significance levels for multiple testing. Standard Bonferroni corrections are too crude since tests at nearby positions are very correlated. In this talk, I describe a method based on extreme value theory for Gaussian processes, which generalizes previous work by e.g. Kruglyak and Lander (1995). The score function (test statistic) is viewed as a stochastic process, with chromosomal position as `time index'. Two important steps of the methods are 1) to calculate the slope of the covariance function at zero and 2) to correct for non-Gaussianity by means of a transformation.