Above and beyond state-of-the-art approaches to investigate sequence data: summary of methods and results from the population-based association group at the Genetic Analysis Workshop 19

Abstract

This paper summarizes the contributions from the Population-Based Association group at the Genetic Analysis Workshop 19. It provides an overview of the new statistical approaches tried out by group members in order to take best advantage of population-based sequence data.Although contributions were highly heterogeneous regarding the applied quality control criteria and the number of investigated variants, several technical issues were identified, leading to practical recommendations. Preliminary analyses revealed that Hurdle-negative binomial regression is a promising approach to investigate the distribution of allele counts instead of called genotypes from sequence data. Convergence problems, however, limited the use of this approach, creating a technical challenge shared by environment-stratified models used to investigate rare variant-environment interactions, as well as by rare variant haplotype analyses using well-established public software. Estimates of relatedness and population structure strongly depended on the allele frequency of selected variants for inference. Another practical recommendation was that dissenting probability values from standard and small-sample tests of a particular hypothesis may reflect a lack of validity of large-sample approximations. Novel statistical approaches that integrate evolutionary information showed some advantage to detect weak genetic signals, and Bayesian adjustment for confounding was able to efficiently estimate causal genetic effects. Haplotype association methods may constitute a valuable complement of collapsing approaches for sequence data. This paper reports on the experience of members of the Population-Based Association group with several novel, promising approaches to preprocessing and analyzing sequence data, and to following up identified association signals.

Omics-squared: human genomic, transcriptomic and phenotypic data for genetic analysis workshop 19

“…At the workshop, investigators used these data to address a wide variety of topics. Analytical issues addressed included methods for population- [14] and family-based [15] association, machine learning and data mining approaches to gene localization [16], and methods for joint analysis of mutiple phenotypes [17]. Some groups concentrated on approaches to dealing with multiple testing in these high dimensional sequence data by filtering sequence variants or placing informative priors for association analyses [18], by pathway-based approaches for gene localization [19], or by other variant collapsing approaches [20].…”