P-Values or Prediction Errors: What Is the Name of the Game in Genomics?

• Activity Number: 599
• Type: Invited
• Date/Time: Thursday, August 5, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Rafal Kustra*+
• Companies: University of Toronto
• Keywords: statistical genomics ; Receiver Operating Curve ; Multiple Testing Problem ; Sensitivity

Abstract

In this talk we consider a (now) ubiquitous multiple testing problem (MTP) of genomics. In majority of cases the genomics experiment is considered an exploratory step that: 1) Attempts to confirm the existance of some correlation between genetic and phenotypic effetcs and 2) Suggest a further experimental roadmap. Most methods today return a sea of p-values (even if adjusted for MTP), which is rarely helpful with the two questions above. Moreover individual p-values which focus on association between single genetic effects and a common phenotype often hide the underlying complexities of the experiment and data analysis - many of which are unevenly spread across the genetic probes. In this talk we will consider whether a paradigm shift in high-throughput genomic analysis may be needed when the number of entities examined grows ever larger with technological advances.