New version of the hglm package available
Together with my colleagues I have developed the R package hglm which fits Hierarchical Generalized Linear Models, i.e. generalized linear models with random effects. See Andrew Gelman's review here.
Scientific Outline
I am a researcher and lecturer in the Statistics group at Dalarna University since April 2008. I am also a guest lecturer at the Dep. of Animal Breeding and Genetics at SLU, Uppsala.
I was a postdoctoral researcher in Carlborg's group at the Linnaeus Centre for Bioinformatics April 2005 till March 2008.
I defended my PhD in March 2003, at the Swedish University of Agricultural Sciences, Uppsala, on the subject "Selection, Maternal Effects and Inbreeding in Reindeer Husbandry". My thesis includes theoretical and applied papers in both ecology and animal genetics. Part of my research was made in collaboration with prof. J.A. Woolliams at Roslin Institute, Edinburgh. After my PhD I was appointed Head of Statistics at Dalarna University and I was responsible for the development of a new Master Program in Statistics . During this time (2003-2005) I collaborated with the research department at Sveriges Riksbank, which resulted in a novel paper on capital risk assessment.
I am currently working in a research project aimed at development of new statistical tools for detection of quantitative trait loci (QTL). The project is funded by FORMAS and involves researchers at Dalarna University, Swedish University of Agricultural Sciences (Örjan Carlborg) and Information Technology at Uppsala University.
The aim of my research is to understand the genetic architecture of quantitative traits. I develop statistical methods for detecting genes in both crosses between divergent experimental lines and natural populations. Divergent experimental crosses are designed to primarily study genetic variation between lines. An important question, that has not been answered yet, is how much variation there is within non-inbred experimental lines and how this may affect our conclusions. In the analysis of natural populations it is assumed that the founders of a pedigree have alleles that have been randomly sampled from a larger population of alleles. What happens if these alleles have not been randomly sampled? Can we develop methods for detecting founder alleles that have not been randomly sampled?
We need to consider epistatic effects to fully understand the genetic architecture of quantitative traits. Methods for detecting genes in general pedigrees are based on variance component estimation. So, to be able to test for many possible gene interaction effects, we need fast variance component estimation methods with highest possible power. Evaluating methods of variance component estimation is therefore also an important part of my research.
I am involved in several PhD-projects in genetics, numerical analysis and statistics.
Since 2009 I have five ongoing research projects focused on hierarchical generalized
linear models (HGLM) and variance component (VC) estimation in genetics.
PhD Project 1: Variance component modeling in QTL analysis
My student Xia Shen is working on two problems:
Including the uncertainty of IBD estimation into the REML likelihood, ii) Using
an HGLM approach to fit VC models for binary traits.
Collaborator:
prof. O. Carlborg, Swedish University of Agricultural
Sciences, Uppsala
PhD
Project 2: Genetic variability of the residual variance
Majbritt
Felleki is studying the potential use of double HGLM in animal breeding
applications
Collaborators:
prof. E Strandberg and Dr.
F. Fikse, Swedish University of Agricultural Sciences, Uppsala
PhD Project 3:
Non-iterative variance component estimation
Razaw
al-Sarraj has developed a new variance component estimation method for linear
mixed models, which is a development of Henderson’s (1953) method III. The
method is currently being evaluated using simulations and real data.
Research
Project: Detecting variance-controlling QTL
The aim
of this project is to develop a regression method for detection of QTL that
regulate the residual variance of a trait. The method is currently being tested
on F2 intercrosses and heterogeneous stocks.
Collaborators: Dr. W. Valdar, Oxford University
Computer
Software Project: HGLM R package
In this
project I am developing an R package for HGLM estimation together with X. Shen
and M. Alam at Dalarna University. The HGLM package is now available
on CRAN.
