Harold Woodruff Lecture Theatre (Old Microbiology 127), Royal Parade, University of Melbourne.
The Centre is a new interdisciplinary initiative involving the science, medicine and engineering faculties and with the support of the DVC-R, which is expected to be fully operational later this year, initially housed in the Microbiology building on Royal Parade. The Centre will focus on all aspects of the genotype to phenotype relationship, including the effects of microbe/pathogen genomes. It will employ advanced maths/stats/computing methods implemented through substantial new computing resources, applied to large volumes of sequencing and other multi-omics data, some of it generated in an integrative genomic profiling lab within the Centre.
We will say a few words about current plans for the Centre before the seminar, and its four founding PIs (David Balding, Edmund Crampin, Kat Holt, Mike Inouye) will be available to answer questions after the seminar.
Guest speaker – Doug Speed
Our guest speaker will be Dr Doug Speed, MRC Biostatistics Fellow at the UCL Genetics Institute, London and author of the freely-available LDAK software which implements a range of tools for SNP-based heritability estimation, phenotypic prediction and gene- or region-based testing, including meta-analysis. Doug will be visiting Melbourne for a further month after the seminar and will also present a different (but related) seminar “Fun ways to analyse genetic data using mixed models” to the Victorian Centre for Biostatistics at 9:30am on Thursday April 30 (room 515, 207 Bouverie St).
Complex trait genetics; the truth is out there … but how do we find it?
Abstract: We now know that common genetic variation explains much of the heritability for most complex traits, and also that polygeneity is the norm; typically, heritability is spread across 100s, if not 1000s of variants, most of very small effect. Through our software LDAK, we are contributing to the current rapid pace of developments of new methods to characterise, identify and exploit causal variants underlying complex traits. Rather than focus on genome-wide significance of a few SNPs, the new methods use large numbers of SNPs (and perhaps all available SNPs) in very large regression models; an approach now feasible thanks to mathematical and computational developments.
In this seminar I will describe some of our recent work. By comparing the heritability of genome regions (or other classes of variants) with their total genetic variability, we can measure the “intensity” of heritability and quantify the relative roles of, for example, exons, eQTLs, and evolutionarily conserved SNPs, finding very different patterns for different traits. We have also developed a fast and powerful method for testing genes or other SNP-sets for influence on the trait. The method can make use of available prior information, and it can be employed across cohorts in a meta-analysis, conveying substantial advantages over traditional single-SNP meta-analysis. MultiBLUP is LDAK’s phenotypic prediction tool which I will show to equal or better leading prediction tools, with greatly reduced computational cost, allowing the tool to be tuned using tens of thousands of training individuals.