Statistical Genetics

Our group historically focused on the quantitative analysis of genetic data and transcriptomic data in psychiatric disease (mostly major depression, but also PTSD and schizophrenia) and its treatment. More recently, we have started to also work on metabolomic as well as proteomic data in the same spirit. Furthermore, we have an interest in other phenotypes, such as dyslexia/dyscalculia, cardiovascular disease, and multiple sclerosis, to name some examples.

We have been developing (statistical and informatics) methods to further our research and enable deeper analysis. To this end, we are trying to marry statistical genetics and machine learning methods, extending the scope of analysis beyond pure genetics. As a consequence of the heavy computational loads many of these methods are facing, we are very interested in algorithmic and generally in computational advances, e.g. OpenCL, and GPGPU (General Purpose Computation on Graphics Processing Unit) implementations on the one hand, and faster algorithms, even approximate or two-stage designs, on the other hand.

We are very much interested in further development of these techniques and also aim to take these approaches further extending from univariate to bivariate and trivariate analyses (such as the epistasis analyses shown in Fig. 1) to network-based approaches, with the aim of still maintaining control of and information about statistical properties of also the components of these networks.

Fig.1: Computational speed of a GPGPU implementation of a test of epistasis. Interaction between two polymorphic variants as compared to more traditional CPU-based implementations, showing that it is easily possible to analyze millions of interactions per seconds in thousands of patients when employing GPGPU technology. [graph is from Kam-Thong et al, Hum Hered, 2012]
Go to Editor View