|
Abstract:
|
Testing SNPs for use as potential biomarkers in drug development has two subtle challenges. The first is that since the drug development process typically involves comparing a new treatment with a control, even if a SNP has a traditional effect in each of the two arms, its effect on differential efficacy may be more complex. Statistical tests currently popular in genetics in fact do not control familywise error rate (FWER) for the effects they infer. The second is that any linkage disequilibrium between non-causal SNPs with a causal SNP renders the zero-null hypotheses of no-association false, making Type I error rate control difficult to interpret. To meet such challenges, we propose a simultaneous confidence interval method based on pivotal statistics. Within each SNP, the confidence intervals we provide are capable of inferring complex efficacy effects while strongly controlling FWER. Across the SNPs, our method controls the expected number of SNPs with false confidence interval coverages. Since our confident effects method infers on effect size, not merely association,it guides the validation process toward SNPs that appear to have the most clinically meaningful effects.
|
