Title: Integrating competing but complementary tests with applications to genetic association studies

Speaker: Integrating competing but complementary tests

Speaker Link: http://www.utstat.toronto.edu/sun/

Affiliation: University of Toronto

Host: Lin Chen

Contact Info: Professor Department of Statistical Sciences Faculty of Arts and Science Division of Biostatistics Dalla Lana School of Public Health

Location: AMB W-229

Start: 4/12/2017 3:30:00 PM

Url: http://health.bsd.uchicago.edu/

Speaker: Integrating competing but complementary tests

Speaker Link: http://www.utstat.toronto.edu/sun/

Affiliation: University of Toronto

Host: Lin Chen

Contact Info: Professor Department of Statistical Sciences Faculty of Arts and Science Division of Biostatistics Dalla Lana School of Public Health

Location: AMB W-229

Start: 4/12/2017 3:30:00 PM

Url: http://health.bsd.uchicago.edu/

Abstract:

In many scientific studies, different statistical tests are being proposed with competing claims about the performance in terms of power. The power of a given test depends on the nature of the alternatives. For example, in the phenotype-genotype association
analyses of complex human traits, the class of location-tests is commonly used to detect phenotypic mean differences between genotype groups. However, complex genetic etiologies including GxG and GxE interactions can result in homoscedasticity, where the
class of scale-tests would be more powerful. In another example where association between multiple genetic variants and a trait is of interest, we show that many existing methods can be classified into a class of linear statistics and another class of quadratic
statistics, where each class is powerful only in part of the high-dimensional parameter space. To achieve robustness it is natural to combine the evidence for association from the two (or more) complementary tests, but how? The well-known Fisher’s method,
commonly used in meta-analyses, combines p-values of the *same *test applied to K
*independent* samples. Here we propose to use it to combine p-values of *
different *tests applied to the *same* sample. In both settings, we show that the two classes of tests are asymptotically independent of each other under the global null hypothesis. Thus, we can evaluate the significance of the resulting Fisher’s
test statistic using the chi-squared distribution with four degrees of freedom; this is a desirable feature for analyzing big data. In addition to theoretical results, we also provide empirical results from extensive simulation studies and multiple data applications
to show that the new class of joint test is not only robust but can also have better power than the individual tests. This is based on work from Derkach et al. (2014)
*Statistical Science*, Soave et al. (2015) The American Journal of Human Genetics, and Soave and Sun (in press)*, Biometrics*.