The effects of climate change
Norwegian Research Council project. Nordic Council Project. (sfi)2 project. The aim is to develop new statistical methods for climate research, with special emphasis on effect studies. We collaborate with Met.no and internationally to understand if global circulation models can be downscaled to the scale that is required for adaptation studies. In cooperation with Norsk Regnesentral, Met.no, Univ. of Washington.
Genomic signature of the efficacy of radiotherapy in breast cancer
Norwegian Research Council project. The aim is to discover genes that interact with radiotherapy and are associated to metastasis free survival after breast cancer surgery. If we find such a gene signature, it might be possible to devise a chip useful to help deciding about radiotherapy. In cooperation with Radiumhospitalet (OUS) and Aarhus University Hospital.
The Genomic HyperBrowser
(sfi)2 project. This project will produce a new open-source, web-based, real time processing, inferential machine, the Genomic HyperBrowser (GHB), for the first time allowing sequence-level, multivariate, local genomic analysis. The GHB represents a radically innovative bridge between biological investigations, complex stochastic models, statistical inference, and computational science. Sequence-level genomic data consist of annotated pieces of information positioned on locations along the genome. There are thousands of such tracks.
We focus on inferential comparison between tracks, to find local significant aberrations from the null-model behaviour. The output is a context-specific track with levels of significance along the genome. The GHB builds on a mathematical representation of tracks, using stochastic geometry. We translate genomic elements into mathematical objects and biological hypotheses become stochastic relations. The analysis concerns stochastic properties of random processes.
Inference corresponds to the computation of probabilities of rare events. The GHB is self-instructing, to integrate data and analyses not yet known. Novel visualisation techniques of results are used, incl. Google.maps technology.
Pair-copula construction of multiple dependence
(sfi)2 project. Multivariate data often exhibit complex patterns of dependence - especially but not solely in the tails - that are difficult to model. We have developed a new class of stochastic models and methods for inference for them, which are a generic tool for multivariate analysis.The pair-copula construction is hierarchical in nature, but the building blocs are very simple: pair-copulae, acting on two variables at a time.
The various levels of the hierarchy correspond to the incorporation of more variables in the conditioning sets. We introduced the modelling framework, discussed inference and properties of estimators. Our applications are in various multivariate situations, and particularly in finance.
Our project will apply pair copula construction to a health related data set, before we will be able to offer our approach to more research groups. This is joint work with Norsk Regnesentral.
New approaches to p>>n regression via penalisation
(sfi)2 project. In a number of important situations, the number of unknown parameters (p) is much larger than the number of samples/observations (n). This is common in genomics, where millions of SNPs are measured on a few hundred individuals. Multivariate modelling and variable selection problems are solved by means of penalisation techniques, like the lasso, which assumes that there are in fact just a few relevant parameters for a certain outcome. This assumption of sparsity is un-structured and non-informative. However, in many situations, there is more knowledge that can be exploited to guide variable selection to deliver more robust prediction rules. In collaboration with the Department of Mathematics at UiO, OUS and Imperial College.
Genomic signature associated to disease and treatment
Various projects. In collaboration with scientific groups at the faculty of medicine and Oslo University Hospital, we are providing statistical genomics expertise to help identify the effects of genes (their expressions, copy number variation, SNPs, etc) on disease and therapy. Currently we are involved in projects concerning
- Breast cancer
- Schizophrenia and other psychotic disorders
We have expertise in genomewide association studies, microarray expressions, copy number variation and sequence based studies.
Modelling antibiotic resistance
HSØ and (sfi)2 project. Modelling antibiotic resistance and alterations in the bacterial flora affected by antibiotic consumption in hospitals. Data are from Oslo region, so an area with low incidence. The challenge is to see if we still can find a sign of causal relation between use of antibiotics and resistance, at a certain lag of time.
A ward specific model approach based on generalised linear models with random intercept and autoregressive errors is proposed. Data are still being collected, and an analysis will be soon finalised. In collaboration with Oslo University Hospital (RH, Aker) and Bærum Hospital.
Genetic influence on brain phenotypes
Norwegian Research Council project. The overall aim of the current research program is to gain more knowledge about pathophysiological mechanisms of severe mental disorders. We explore gene – brain relationships and how this is related to clinical characteristics and neurocognitive function.
Our partners take advantage of their access to international genetic consortia with huge samples of genotyped cases and controls, existing datasets with genetic, brain imaging and neurocognitive information, and they recruit and characterize new cohorts with new equipment and technology in focused studies. In particular we work with fMRI data and try to associate features of these with genomic information. In collaboration with the TOP group at Ullevåll, Oslo University Hospital.
(sfi)2 project. Retroviral vectors are widely used in gene therapy to introduce therapeutic genes into the cells of patients: the transported genes are stably integrated into the target DNA. There is now compelling evidence that integration of retroviral vectors follows non-random patterns, with a preference for active genes and regulatory regions. Using extensive integration data, we are comparing integration preferences of various viruses, in particular the Moloney Leukemia Virus (MLV) and the HIV. Statistically significant differences in patterns of integration will allow choosing the best carrier. In collaboration with Universitá La Salute - San Raffaele, Milano.
Stochastic models of infectious diseases in fish farming
(sfi)2 project. Infectious diseases are a constant threat to the fish farming industry with major economic implications, in addition to being a problem for fish welfare and the environment. The occurrence of salmon lice in Norwegian fish farms was exploding in the summer and autumn 2009, and is currently the largest health problem in Norwegian fish farming. Infectious salmon anaemia (ISA) has over years been one of the most dangerous infectious diseases in several countries.
The spread of infectious diseases, as a complex dynamical system with high level interaction, has an inherent element of randomness. We develop spatio-temporal stochastic models to describe routes of dissemination of a disease between fish farms, for testing hypotheses on transmission pathways and for quantifying the effects of potential preventive actions. In collaboration with Veterinærinstituttet and Norsk Regnesentral.