A LASSO penalized regression approach for genome-wide association analyses using related individuals: application to the Genetic Analysis Workshop 19 simulated data

Abstract

We propose a novel LASSO (least absolute shrinkage and selection operator) penalized regression method used to analyze samples consisting of (potentially) related individuals. Developed in the context of linear mixed models, our method models the relatedness of individuals in the sample through a random effect whose covariance structure is a linear function of known matrices with elements combinations of the condensed coefficients of identity between the individuals in the sample. We implement our method to analyze the simulated family data provided by the 19th Genetic Analysis Workshop in an effort to identify loci regulating the simulated trait of systolic blood pressure. The analyses were performed with full knowledge of the simulation model. Our findings demonstrate that we can significantly reduce the rate of false positive signals by incorporating the relatedness of the study participants.

Family-based approaches: design, imputation, analysis, and beyond

“…The remaining contributions focused on diverse aspects of analysis of the trait data in family-based samples. Two papers by Papachristou et al [ 34 ] and Zhou et al [ 35 ] addressed computational challenges of carrying out family-based analysis of multivariate data by using a 2-stage approach: a rapid initial analysis that ignores relatedness, followed by a computationally more challenging analysis to correct for the effects of related subjects among the loci implicated in the first stage. The paper by Papachristou et al [ 34 ] focused on carrying out tests among multiple, correlated markers, while that by Zhou et al [ 35 ] focused on tests carried out among multiple, correlated, traits.…”

“…Two papers by Papachristou et al [ 34 ] and Zhou et al [ 35 ] addressed computational challenges of carrying out family-based analysis of multivariate data by using a 2-stage approach: a rapid initial analysis that ignores relatedness, followed by a computationally more challenging analysis to correct for the effects of related subjects among the loci implicated in the first stage. The paper by Papachristou et al [ 34 ] focused on carrying out tests among multiple, correlated markers, while that by Zhou et al [ 35 ] focused on tests carried out among multiple, correlated, traits. Two papers carried out association testing only with variants of pedigree-based approaches implemented in FBAT (family-based association test) [ 36 ] or a similar approach [ 37 ] that condition on possible transmission of alleles or genotypes in a joint test of association in the presence of linkage.…”

“…Darst and Englelman [ 41 ] compared performance of the pedigree-based FBAT to a linear mixed model approach to correct for correlation for relatedness. Three other papers also applied a similar linear mixed model approach to the family data [ 31 , 34 , 35 ]. Finally, Saad et al [ 32 ] introduced a novel approach to combine transmission information obtained from pedigree data with identity-by-descent (IBD) inferred between individuals in different pedigrees to increase linkage evidence, thus, in some senses, combining both pedigree- and family-based approaches.…”

Strategies for integrated analysis in imaging genetics studies

“…intrinsically performs feature selection and regularization, and shrinking coefficients of non-selected predictors towards zero (Tibshirani, 2011). LASSO is commonly used in genetic studies (Papachristou, Ober, & Abney, 2016;Pineda et al, 2014), and has also been used in IG (Kohannim et al, 2012).…”

L1 Penalized Cox Regression to Characterize Cardiovascular Events in Hypertensive Patients

“…By augmenting the fitting regression term with a L 1 -norm cost, LASSO simultaneously performs regularization (leading to a more generalizable model) and feature selection (enforcing sparsity in the final solution). This approach has been also applied in genome-wide association analysis [5] and when trying to identify prognostic factors with high-dimensional data such as radiological features of PET images [6] or environmental enteropathy biomarkers [7]. In these situations, traditional statistical methods for feature selection may be tedious or inefficient due to the amount of covariates and the non-obvious correlation among them.…”