MMAP: a comprehensive mixed model program for analysis of pedigree and population data. J. O'Connell Div Endo/Diabetes/Nutrition, Univ Maryland, Baltimore, MD.

   The application of mixed models for genetic analysis has seen a sharp increase the last several years. The power and flexibility of the mixed model has been applied to both population and pedigree data, to both single SNP and multi-SNP data as random and fixed effects, to gene expression and general omics data, to both estimation and prediction of genetic risk, to mapping both common and rare variants, and to understanding the genetic architecture of complex traits. Kernel machine methods, kriging, and non-parametric regression can be cast into the mixed model framework to take advantage of available machinery. The basic ingredients in the mixed model are covariance matrices that measure similarity between subjects based on genetic data.
   MMAP is an optimized and flexible mixed model analysis platform that incorporates a wide range of covariance structures such as additive, dominance, epistasis, maternal and imprinting using pedigree and/or genotype data and also allows users to define their own covariance structures. Likelihood calculations use multi-threaded optimized matrix libraries to handle multiple random effects. MMAP can import data from a variety of imputation programs to avoid data manipulation and IBS/IBD programs to build covariance structures.
   MMAP uses a fast low-memory method to calculate additive and dominant genetic covariance structures using SNP data, which can be quite challenging for large data sets. For polygenic SNP analysis MMAP can store SNP-covariance products to reduce the complexity subsequent analyses with the same subjects to linear regression, independent of phenotype or covariates.
   We present results in both animal and human data. We show that the genetic architecture of complex traits can include significant non-additive variance such as dominance in milk traits estimated in 35,000 Holstein cows and in blood pressure in 3500 Amish subjects. We show that including sources of non-additive variance in estimating SNP effects can improve predictive ability. We also present a detailed comparison between pedigree and genomic estimates of heritability in the Amish to measure missing heritability.

You may contact the first author (during and after the meeting) at