|
Abstract:
|
Single nucleotide polymorphisms (SNPs) may help predict patient response to drug treatment. In drug development, it is more desirable to identify the the single, best predictive SNP. To that end, most proposed procedures are based on the p-value ordering. However, the p-value ordering does not necessarily correlate with the true ordering. To address this problem, in our paper, we propose a rank-based simultaneous-confidence-intervals approach in the formulation of multiple comparisons with the best. One of the main contributions of our research is that we formulate the parameters of interest to compare each candidate SNP with the others in terms of their predictive effects directly. We then build lower-bounded simultaneous confidence intervals in corporate with linkage disequilibrium to quantify the parameters of interest. A new error rate, the expected number of false coverage, is controlled. Top SNPs are selected based on either lower bounds rank or minimal clinically important difference. We compared our method with other commonly used methods, including some machine learning approaches. It turns out that our framework has a better chance to give the correct assertions.
|
