Multiple SNP analysis with HyperLasso in Pharmacogenetic Studies using Nonlinear Mixed Effects Models
Julie Bertrand (1, 2), David Balding (2)
(1) UMR738 INSERM University Paris Diderot, Paris, France; (2) Genetic Institute of the University College of London, London, England
Context: To date the influence of genetic polymorphisms on concentration data is usually analyzed using the observed area under-the curve, maximal or minimal concentrations. Nonlinear mixed effects models (NLMEM) enable the analysis of all concentration versus time profiles, even on sparse data. This allows more flexibility in study design, for example to include additional subjects with fewer samples, making it easier for pharmacogenetic studies to meet the requirements of the medical authorities [1]. Lehr et al. [2] provide the first evaluation of the feasibility of, and potential benefits from, integrating multiple SNP data into NLMEM. However, their algorithm is a stepwise procedure which is known to have problems with correlated covariates. The HyperLasso approach [3] simultaneously analyses all SNPs using a prior with a sharp peak at zero and heavy tails.
Objective: To assess the type I error and power of HyperLasso and stepwise-based approaches for detecting a SNP effect on a pharmacokinetic parameter using NLMEM.
Methods: Two hundred data sets were simulated under both the null and an alternative hypothesis. Genetic polymorphisms were simulated using HAPGEN [4] approximating the design of the DMET chip [5] with about 1200 genetic polymorphisms across the all genome. Pharmacokinetic profiles were simulated using a two-compartment model with parameters based on real data [6]. Under the alternative hypothesis three causal variants were randomly associated to one or several PK parameters with a rare allele dosage effect size of 10 to 20%, so that the impact of the allele frequency and effect size could be investigated. PK modeling was performed using the Stochastic Approximation Expectation Maximisation algorithm implemented in Monolix 3.1.
Results: In average 830 over the 1252 genetic covariates were polymorphic in the population of 300 subjects per simulated data set. Under the null hypothesis of no gene effect in average [range], 4.5 [2-5] and 1.4 [0-8] polymorphisms were found associated to one or more of the 5 PK model parameters using the HyperLasso method on Empirical Bayes Estimates and a forward inclusion in the model, respectively.
References:
[1] EMEA, EMEA: Reflection paper on the use of pharmacogenetics in the pharmacokinetic evaluation of medicinal products, 2007.
[2] T. Lehr, H.-G. Schaefer, and A. Staab, "Integration of high-throughput genotyping data into pharmacometric analyses using nonlinear mixed effects modeling.," Pharmacogenetics and Genomics, vol. 20, 2010, pp. 442-450.
[3] C.J. Hoggart, J.C. Whittaker, M. De Iorio, and D.J. Balding, "Simultaneous Analysis of All SNPs in Genome-Wide and Re-Sequencing Association Studies," PLoS Genetics, vol. 4, 2008, p. 8.
[4] J. Marchini, B. Howie, S. Myers, G. McVean, and P. Donnelly, "A new multipoint method for genome-wide association studies by imputation of genotypes.," Nature Genetics, vol. 39, 2007, pp. 906-913.
[5] T.M. Daly, C.M. Dumaual, X. Miao, M.W. Farmen, R.K. Njau, D.-J. Fu, N.L. Bauer, S. Close, N. Watanabe, C. Bruckner, P. Hardenbol, and R.D. Hockett, "Multiplex assay for comprehensive genotyping of genes involved in drug metabolism, excretion, and transport.," Clinical Chemistry, vol. 53, 2007, pp. 1222-1230.
[6] B.S. Kappelhoff, A.D.R. Huitema, Z. Yalvaç, J.M. Prins, J.W. Mulder, P.L. Meenhorst, and J.H. Beijnen, "Population pharmacokinetics of efavirenz in an unselected cohort of HIV-1-infected individuals.," Clinical pharmacokinetics, vol. 44, Jan. 2005, pp. 849-61.
