unisa logo

Australian Centre for Precision Health

Population health research to improve the future of health

Puzzle of Genes

Developing statistical methods for whole genome analysis

In the genomic era, there are numerous genotype and phenotype data publicly available as a form of ‘big genomic data’. Genome-wide genotype information has provided valuable insights into the genetic basis of complex human diseases. It is now increasingly recognised that whole-genome approach is useful in complex disease analyses, which can use all or most genetic variants across the genome simultaneously. The approach is to link two individuals who are not related in the conventional sense, but who can be compared experimentally because they share part of their genome by descent over many generations. This is a paradigm-shifting approach, leading to a design-free experiment for population genetic analyses that does not require pedigree-informative individuals or relatives. Combined with advanced statistical methods, the whole-genome approach is a promising tool to dissect the genetic architecture and maximise the accuracy of risk prediction for complex diseases, leading to effective precision medicine.

We are currently developing advanced whole-genome methods for causative variant detection, genotype-environment interaction and dissection of a dynamic genetic architecture of complex traits to maximise the accuracy of individual risk prediction.

Available software

MTG2 is a computer program implementing a multivariate linear mixed model to fit complex covariance structures that can be constructed based on genomic information, i.e. multivariate version of GCTA REML. It gives residual maximum likelihood (REML) estimates for genetic and environmental variance and covariance across multiple traits. It estimates the best liner unbiased prediction (BLUP) for quantifying genetic merits or genetic risk. MTG uses the direct average information algorithm. Recently, we combined the direct AI algorithm with an eigen-decomposition of the genomic relationship matrix, as first proposed by Thompson and Shaw (1990). We apply the procedure to analyse real data with univariate, multivariate and random regression linear mixed models with a single genetic covariance structure, and demonstrate that the computation efficiency can increase by >1,000 fold compared with standard REML software based on MME. In addition, random regression models and reaction norm models are available for univariate and multivariate frameworks. There are many other functions in complex trait analyses and statistical genetics (see contents table in the manual). 

Access MTG2 here

Current research projects