New Zealand Statistical Association

NZSA 2009

Victoria University of Wellington

Rod Ball

Scion

A Virtual Institute of Statistical Genetics

The Virtual Institute of Statistical Genetics (VISG) is a FRST funded research program involving Universities and Crown Research Institutes, that has just completed its first year of operation. I will describe progress on current projects (Large Datasets and Polyploids) and plans for the next project (Experimental Designs).

Whole genome prediction of genetic values is already being applied to livestock in New Zealand. The VISG large datasets project is developing methods for whole genome association mapping and prediction of genetic values. Currently we are working on a Bayesian Markov chain Monte Carlo (MCMC) method for fitting associations using a multi-category prior (with 3 categories, with separate variance parameters for each category) and block updates for SNP effects. The modelling approach allows for low prior probabilities for non-negligible SNP effects (necessary for a p << n problem), and a non-normal mixture distribution for effect sizes. Special attention is given to algorithms for improving and diagnosing MCMC convergence to avoid problems with existing QTL or whole genome MCMC methods. Also important are algorithms to effectively handle large datasets with currently of the order of 1,000,000 SNP markers genotyped per individual.

A number of important horticultural, crop, and forage species are polyploids. Existing QTL mapping in polyploids is limited to specific marker types and segregation patterns, and inference is limited. The VISG polyploids project is developing methods for QTL mapping in polyploids which make full use of available marker information and enable multi-locus Bayesian inference of the genetic architecture.

Polyploids have 2 or more sub-genomes resulting in (e.g. for an allo-tetraploid) 4 or more alleles at each locus each of which could have been inherited from one of 8 grand-parental chromosomes. Markers are rarely fully informative, so that the statistical method needs to contend with considerable and variable amounts of missing information. This is being done by integrating peeling and conditional peeling with a Bayesian QTL mapping method.

Experimental design has been a neglected area in genomics, with even large scale international projects lacking power to detect any but the largest effects with posterior odds greater than 1. The VISG experimental designs project will develop experimental designs with sufficient power to detect genomic associations, with sufficiently high Bayes factor to overcome the low prior odds for genomic associations, and utilising design and analysis options available in various species (e.g. clonal replication and spatial analysis).
Contact Us | Section Map | Disclaimer | RSS feed RSS FeedBack to top ^

Valid XHTML and CSS | Built on Foswiki

Page Updated: 30 Aug 2009 by haywoodj. © Victoria University of Wellington, New Zealand, unless otherwise stated. Header image used and relicensed under Creative Commons. Original author: Djof.