The Power to Detect Linkage in Complex Disease by Means of Simple LOD-Score Analyses

Abstract

Maximum-likelihood analysis (via LOD score) provides the most powerful method for finding linkage when the mode of inheritance (MOI) is known. However, because one must assume an MOI, the application of LOD-score analysis to complex disease has been questioned. Although it is known that one can legitimately maximize the maximum LOD score with respect to genetic parameters, this approach raises three concerns: (1) multiple testing, (2) effect on power to detect linkage, and (3) adequacy of the approximate MOI for the true MOI. We evaluated the power of LOD scores to detect linkage when the true MOI was complex but a LOD score analysis assumed simple models. We simulated data from 14 different genetic models, including dominant and recessive at high (80%) and low (20%) penetrances, intermediate models, and several additive two-locus models. We calculated LOD scores by assuming two simple models, dominant and recessive, each with 50% penetrance, then took the higher of the two LOD scores as the raw test statistic and corrected for multiple tests. We call this test statistic "MMLS-C." We found that the ELODs for MMLS-C are >=80% of the ELOD under the true model when the ELOD for the true model is >=3. Similarly, the power to reach a given LOD score was usually >=80% that of the true model, when the power under the true model was >=60%. These results underscore that a critical factor in LOD-score analysis is the MOI at the linked locus, not that of the disease or trait per se. Thus, a limited set of simple genetic models in LOD-score analysis can work well in testing for linkage.

Parametric and nonparametric genome scan analyses for human handedness

“…However, it has been suggested that twopoint analysis with a simple genetic model with reduced penetrance has the power to detect linkage for a trait with a complex mode of inheritance. 21 The reduced penetrance makes the model robust enough to detect linkage and it mimics the effect of multiple loci. 21 In addition, the high gene frequency specified in the model of Klar makes the model more robust as it allows for the parents passing on either haplotype if they are homozygous for the trait locus.…”

A full genome screening in a large Tunisian family affected with thyroid autoimmune disorders

“…Linkage analysis was performed using the LINKAGE package, 27 under two arbitrary models, one dominant (disease allele frequency: 0.01; penetrance in heterozygotes and homozygotes: 0.5; phenocopy rate: 0.0), and one recessive (disease allele frequency: 0.01; penetrance in homozygotes: 0.5; phenocopy rate: 0.0), for a recombination fraction = 0.0. For each marker, a corrected parametric linkage score depending on these two results (MMLS-c) was obtained as the maximum of both lod score values, reduced by 0.30 according to a method proposed by Greenberg et al 28 Fine exploration of chromosome 2 region was performed on a total of 107 individuals, including the first set of 92 individuals used for genome-wide screening. Linkage analysis was performed, using the same statistics as for the genome screening.…”

A genome scan in multigenerational families with dyslexia: Identification of a novel locus on chromosome 2q that contributes to phonological decoding efficiency

“…However, linkage detection with parametric approaches is well-known to be robust to parameter misspecification, 61,70,71 with only modest reduction in power to detect linkage 61,72 even under multilocus inheritance models that are analyzed under single locus models. 73 An inflated trait allele frequency or consequent contribution to the variance is the major expected effect of failure to adjust for ascertainment. 74 This inflated allele frequency may actually be beneficial for robust linkage detection and relatively accurate localization because artificially high trait allele frequencies can compensate for model misspecification in model-based linkage analysis.…”