Weighted functional linear regression models for gene-based association analysis

Nadezhda M Belonogova, Gulnara R Svishcheva, James F Wilson, Harry Campbell, Tatiana I Axenovich

Research output: Contribution to journalArticlepeer-review


Functional linear regression models are effectively used in gene-based association analysis of complex traits. These models combine information about individual genetic variants, taking into account their positions and reducing the influence of noise and/or observation errors. To increase the power of methods, where several differently informative components are combined, weights are introduced to give the advantage to more informative components. Allele-specific weights have been introduced to collapsing and kernel-based approaches to gene-based association analysis. Here we have for the first time introduced weights to functional linear regression models adapted for both independent and family samples. Using data simulated on the basis of GAW17 genotypes and weights defined by allele frequencies via the beta distribution, we demonstrated that type I errors correspond to declared values and that increasing the weights of causal variants allows the power of functional linear models to be increased. We applied the new method to real data on blood pressure from the ORCADES sample. Five of the six known genes with P < 0.1 in at least one analysis had lower P values with weighted models. Moreover, we found an association between diastolic blood pressure and the VMP1 gene (P = 8.18×10-6), when we used a weighted functional model. For this gene, the unweighted functional and weighted kernel-based models had P = 0.004 and 0.006, respectively. The new method has been implemented in the program package FREGAT, which is freely available at https://cran.r-project.org/web/packages/FREGAT/index.html.

Original languageEnglish
Pages (from-to)e0190486
JournalPLoS ONE
Issue number1
Publication statusPublished - 8 Jan 2018


  • Journal Article


Dive into the research topics of 'Weighted functional linear regression models for gene-based association analysis'. Together they form a unique fingerprint.

Cite this