Return to Colloquia & Seminar listing
Identifying Differentially Expressed Genes via Multiscale Geometric AnalysisSpecial Events
|Speaker:||Gilad Lerman, Courant Institute, New York University|
|Start time:||Mon, Jan 26 2004, 4:10PM|
We confront some problems of data analysis (in particular outlier detection with applications to bioinformatics) and use ideas developed in harmonic analysis (specifically stopping-time constructions and multiscale geometric analysis). The first problem is the identification of differentially expressed genes or, more generally, the nonparametric detection of outliers in heteroskedastic data. We begin by assuming the data is normalized so that it is concentrated around a line. We suggest a multiscale construction for a "strip" with varying width around the line. The strip is intended to separate the "deviating" points or genes from the rest. We may generalize to the case where the data is not normalized a priori, so that we construct a strip of varying width around a curve. We also discuss briefly more general constructions, possible applications, difficulties, and relevant geometric information theory developed by Peter Jones and the speaker. This is a joint work with Joe McQuown and Bud Mishra as well as continuing theoretical work with Peter Jones.