Title: On high-dimensional prediction for diagnostics
Authors: Sach Mukherjee - German Center for Neurodegenerative Diseases (Germany)
Bernd Taschler - German Center for Neurodegenerative Diseases (Germany) [presenting]
Abstract: The purpose is to discuss questions that arise in the use of high-dimensional prediction methods in diagnostic applications, illustrated by a case study in leukaemia. We focus, in particular, on machine learning and sparse regression approaches that learn high-dimensional predictive signatures directly from genome-wide data. We discuss issues arising from batch and site effects, the importance of understanding the effect of prevalence in the tested population and on the need for careful empirical assessment of predictors.