Jones Richard, Pembrey Marcus, Golding Jean, Herrick David
ALSPAC, Department of Community-Based Medicine, University of Bristol, Bristol, UK.
Paediatr Perinat Epidemiol. 2005 Jul;19(4):264-75. doi: 10.1111/j.1365-3016.2005.00664.x.
All the approaches to the search for genotype/phenotype associations have their share of problems. Comparing the genome scan and candidate gene approaches, the former makes fewer assumptions at the genetic level or about mechanism but has greater statistical difficulties while the latter partially solves the statistical problem but makes more assumptions at both genetic and mechanistic levels. Among current difficulties is a lack of information about the nature of gene variant/phenotype associations: the frequency with which different classes of gene or sequence are involved; the type of genetic variation most commonly involved; the appropriate genetic models to apply to analysis. The overarching problem is that of multiple testing, one solution to which is to integrate genetic information to create a smaller number of compound variables. At the other end of the scale, decisions about the level of complexity at which to pitch the identification of phenotypes also affect the multiple testing problem: whether to pitch them at the level of disease outcomes, or at any of the multiple levels of intermediate phenotypes or traits. The third issue is how best to deal with gene/gene or gene/environment interactions, or whether to ignore them. Only as more genotype/phenotype associations emerge, by whatever means, will the numbers of results allow these questions to be answered. We describe here a new approach to genotype/phenotype association studies, the phenome scan, in which dense phenotypic information in human cohorts is scanned for associations with individual genetic variants. We believe that this approach can generate data that will be useful in answering generic questions about genotype/phenotype associations as well as in discovering novel ones.
所有寻找基因型/表型关联的方法都有各自的问题。比较全基因组扫描和候选基因方法，前者在遗传水平或机制方面的假设较少，但存在较大的统计困难，而后者部分解决了统计问题，但在遗传和机制层面都做出了更多假设。当前的困难之一是缺乏关于基因变异/表型关联性质的信息：不同类型的基因或序列涉及的频率；最常涉及的遗传变异类型；适用于分析的遗传模型。首要问题是多重检验，一种解决办法是整合遗传信息以创建数量更少的复合变量。另一方面，关于确定表型的复杂程度的决策也会影响多重检验问题：是将其设定在疾病结果层面，还是设定在中间表型或性状的多个层面中的任何一个层面。第三个问题是如何最好地处理基因/基因或基因/环境相互作用，或者是否忽略它们。只有通过各种方式出现更多的基因型/表型关联，结果数量才足以回答这些问题。我们在此描述一种新的基因型/表型关联研究方法——表型组扫描，即对人类队列中的密集表型信息进行扫描，以寻找与个体遗传变异的关联。我们相信这种方法能够生成有助于回答关于基因型/表型关联的一般性问题以及发现新关联的数据。
