Statistical Prioritization of Sequence Variants

• Activity Number: 571
• Type: Invited
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lisa Joanna Strug*+ and Weili Li
• Companies: The Hospital for Sick Children and University of Toronto and The Hospital for Sick Children and University of Toronto
• Keywords: likelihood ratio ; composite hypothesis tests ; next generation sequencing ; sequence analysis ; exact tests

Abstract

Genetic association studies of sequence variants require prioritization for follow-up. Ranking by p-value provides conceptually confusing results. Consider the simple case where disease-variant distributions can be summarized in 2x2 tables, and exact p-values are used to rank variants. Due to the asymmetry of the hypergeometric distribution, rankings by 1-sided and 2-sided p-values can differ. When the distribution is in complete or quasi-complete separation, we show analytically that the LOD score, computed from the ratio of conditional likelihoods, always provides the same rankings as 1-sided p-value with the alternative "correctly" specified. Using simulated data we (1) show that rankings by LOD scores correlate better with the underlying odds ratios, (2) investigate ranking as a function of sample size, and (3) compare the frequencies with which causal variants are prioritized. Variant ranking with LOD scores use only the observed data. P-values incorporate the probability of more extreme data, defined by a priori belief in the direction of association. These findings question the utility of variant ranking using composite hypothesis tests, and suggest the LOD as an alternative