Family-Based Association Test Principal Components (FBAT-PC)

This definition of FBAT-PC is courtesy of Jessica Lasky-Su from the following publication:

Lasky-Su J, Faraone SV, Lange C, Biederman J, Tsuang M, Doyle AE, Smoller J, Laird N, Sklar P. Family based association of statistically derived quantitative traits for ADHD reveal an association in DRD4 with inattentive symptoms in ADHD individuals. American Journal of Medical Genetics, Neuropsychiatric Genetics . 2008 Jan 5;147(1):100-6.24.

Family-based association tests are those studies that use genetic data from family members to evaluate the possible association of a disease phenotype and a gene allele. FBAT-PC is an approach designed to maximize the genetic information when multiple phenotypes are tested (Lange, van Steen et al. 2004) ; it has been implemented in the PBAT computer program (Lange, DeMeo et al. 2004) . The FBAT-PC methodology has been extremely successful in identifying SNPs associated with complex diseases in genomewide association studies. Recently, a 100K scan reported an obesity gene that met genomewide significance using this methodology with replication in several independent samples across various ages and ethnicities(Herbert, Gerry et al. 2006).

When multiple phenotypes are used, FBAT-PC uses principal components analysis to construct an overall phenotype that amplifies the trait heritability by aggregating the genetic components of all measurements into a single univariate phenotype with maximal heritability. This univariate trait with maximal heritability is constructed as follows. First, the conditional mean model proposed by Lange et al. (Lange, DeMeo et al. 2003) is used to generate genetic effect sizes for each phenotype. The effect size estimates are used along with information on the mode of inheritance and allele frequencies to estimate a matrix of genetic variances at that SNP. Second, the phenotypic variance matrix is calculated using the original set of phenotypes as well as any other desired covariates. Once the genetic and phenotypic variance matrices are generated, generalized principal components are used to identify the set of weights that maximize the heritability at the given SNP; heritability is calculated as the weighted combination of genetic variances divided by the weighted combination of the total variance (the sum of the genetic and phenotypic matrices). We apply this methodology to 9 inattentive and 9 hyperactive-impulsive symptoms for ADHD. The inattentive symptoms include: 1) inability to pay attention to details; 2) difficulty with sustained attention in tasks or play activities; 3) listening problems; 4) difficulty following instructions; 5) problems organizing tasks and activities; 6) avoidance or dislike of tasks that require mental effort; 7) tendency to lose things like toys, notebooks, or homework; 8) distractibility; and 9) forgetfulness in daily activities. Hyperactive-impulsivity symptoms include: 1) fidgeting or squirming; 2) difficulty remaining seated; 3) restlessness; 4) difficulty playing quietly; 5) always seeming to be "on the go"; 6) excessive talking; 7) blurting out answers before hearing the full question; 8) difficulty waiting for a turn or in line; and 9) problems with interrupting or intruding. Using only one of these ADHD symptoms does not provide as much information as using a phenotype that is generated from many of the symptoms combined using FBAT-PC.

Once the univariate phenotype is generated at each SNP, FBAT-PC employs a screening procedure that selects the SNPs to be tested using a univariate quantitative FBAT statistic. Such a statistic is called the FBAT-PC statistic. The screening procedure works as follows: 1) for each SNP, power to detect association with the generated univariate phenotype is calculated; 2) a group of SNPs with their associated phenotypes are selected based on the power to detect a genetic association and; 3) the FBAT-PC statistic is calculated on the SNPs and their associated phenotypes. In this analysis, the additive genetic model was used in the screening procedure and a minimum of 20 informative families were required for any given SNP to be screened. The 5 SNP/genetic model combinations with the greatest power were retained and the FBAT-PC statistic was calculated. These five SNPs were subsequently adjusted using the false discovery rate (Benjamini and Hochberg 1995).

The contribution of each ADHD symptom to the final univariate phenotype was determined by first looking at the correlation of each ADHD symptom with the phenotype that was generated from the FBAT-PC program. We then ranked these correlations from the highest to the lowest to determine what symptoms were most strongly correlated with the generated phenotype. If the ranking of the correlations revealed that a clinically meaningful subgroup of ADHD symptoms, FBAT-PC was rerun with the selected group of ADHD symptoms and the p-value from the FBAT-PC statistic was directly evaluated at the SNPs previously found to be significant ( i.e. no screening procedure was used).