In general, large GSVs are more readily detected from genotyping or CGH array data, but occur infrequently so that phenotype associations are difficult to demonstrate statistically. Conversely, accurate calling of smaller GSVs spanning only a few probes remains problematic, despite ongoing improvements in methods for GSV detection [19], [33], [34]. Inaccuracy in GSV calling, with an appreciable frequency of both false positive and false negative calls, results in inflated P-values and an increased rate of false-positive associations (Text S2; Figure S4). This is conceptually equivalent to other scenarios in which genome-wide inflation occurs as a result of genotyping inaccuracy [35] – indeed, attempts to apply algorithmic detection of GSVs to genome-wide genotyping data yield results with marked inflation [36] (our unpublished observations) – and we suggest that the appropriate correction is to apply established methods of genomic control, e.g. scaling of χ2 values according to the genomic inflation parameter λ.
