CAT scans, PET scans, and genomic scans.

As we begin the long march toward genetic dissection of complex traits, it becomes necessary to develop optimum study designs and retool ourselves to face the emerging new challenges. Key issues pertaining to the design of genomic scans are reviewed, including: sampling unit, definition and refinement of phenotype, genotyping issues, one-stage vs. two-stage strategies, sample size and power, and cost and feasibility. It is emphasized that false positives should not be minimized in isolation from the issue of false negatives. Striking a practical balance between the two error rates is suggested. In terms of future directions to pursue, three areas are suggested: meta-analysis for pooling linkage results from multiple scans, rapid multivariate screening methods for increased power to detect quantitative trait loci (QTLs), and classification and regression trees (CART) methodology for handling heterogeneity and interactions. Finally, three recommendations are proposed for genomic scans. First, so as to minimize false negatives for a fixed sample size, it is recommended that we tolerate/accept a reasonable rate of false positives, on average, one false positive per individual scan. Second, so as to enable the use of relatively strict significance levels for interpreting the results from a genomic scan, it is highly recommended that the sample size be derived based on a significance level of at most 0.01 (and not 0.05) and 90% power (and not 80%). Third, it is recommended that the stringent significance levels suggested by Lander and Kruglyak be used when pooling evidence from multiple genomic scans (and not at the level of individual scans).