Identifying useful predictors from genome-wide SNP association studies

使用拥有多达100万个标记的全基因组SNP基因分型技术进行疾病-遗传标记关联研究正变得司空见惯。通常，这类研究的既定目标是开发一种“基因预测因子”，以确定人群中那些罹患疾病风险较高的人。然而，这些研究的统计分析往往包括简单的疾病与每个标志物之间的联系的测试。尽管任何一种与疾病相关的标记物都会有一些预测作用，但这种一次一项的关注过于简单化，原因有几个。首先，如果一个特定的标记变异在人群中很少见，它的预测能力将是有限的，即使相关的风险或优势比相当大。其次，当不同的基因或标记高度相关时，不能通过单独检查每个基因或标记的影响来计算组合的预测能力。第三，一些基因或标记的影响可能是通过与环境暴露的相互作用或与其他基因的相互作用来实现的。也需要考虑这种可能性，尽管随之而来的是统计测试的数量增加。

Disease-genetic marker association studies, using genome-wide SNP genotyping technologies with up to a million markers, are becoming commonplace. Often, the stated goal of such studies is to develop a "genetic predictor" to identify those in the population who are at higher risk of developing disease. However, the statistical analyses of these studies have tended to comprise simple tests for association between disease and each marker in turn. Although any single disease-associated marker will have some predictive utility, this one-at-a-time focus is too simplistic for several reasons. Firstly, if a particular marker variant is rare in the population, it will have limited predictive capability, even if the associated risk or odds ratio is quite large. Secondly, when distinct genes or markers are highly correlated, the predictive capability of the combination cannot be calculated by examining the effects of each separately. Thirdly, the impact of some genes or markers may be through interactions with environmental exposures or interactions with other genes. Such possibilities also need to be considered, despite the concomitant increase in the number of statistical tests.