The Missing Heritability Paradigm: A Dramatic Resurgence of the GIGO Syndrome in Genetics

tions reporting heritability estimates for different multifactorial diseases and claiming the existence of some socalled missing heritability [5] . In these computations of heritability ( table 1 ), the assumed model for multifactorial diseases is no longer a monogenic model but a polygenic additive one with a liability threshold [6–9] . An infinite number of factors (both genetic and environmental) with weak, independent and additive effects are assumed to contribute to an underlying liability that is normally distributed, and a threshold is defined beyond which an individual is affected. Rather than the heritability of a multifactorial disease, it is thus the heritability of the liability of the disease that is estimated. It is also under this model that a novel method has been developed that allows the computation of heritability estimates from genome-wide association study data of unrelated individuals [10, 11] . Is this model suitable for multifactorial diseases? Some argue that observations are consistent with this model [12] : the relative risks estimated at individual SNPs are weak, there is no evidence for statistical interaction between associated SNPs, and the variance explained by each chromosome is proportional to its length. However, consistency with a model does not mean that the model is true. It may simply result from In the search for the genetic factors underlying multifactorial diseases, the way is paved by epidemics of the GIGO (Garbage-In Garbage-Out) syndrome. Indeed, the use of simplistic models for these diseases leads to erroneous conclusions. The first epidemic of the GIGO syndrome occurred in the 1980s with the publication of the first comprehensive genetic maps of markers that enabled linkage analysis at the genome-wide level. During this decade, the lod score method, previously developed by Morton [1] to find genes involved in monogenic diseases, was used in the study of multifactorial diseases. The method requires the specification of a genetic model for the correspondence between the genotypes at the disease locus and the phenotype [2] . For most multifactorial diseases that involve multiple genetic and environmental risk factors, this correspondence is not known. The simple model used for monogenic diseases that assumes a rare, highly penetrant, dominant or recessive gene mutation is irrelevant for multifactorial diseases. This did not prevent researchers from using this model, and numerous tables of meaningless lod scores were published at that time, leading to invalid conclusions [3] . Since the first report of the complete human genome sequence in 2008 [4] , we have witnessed a dramatic resurgence of the GIGO syndrome, with numerous publicaPublished online: February 6, 2015