Predicting functional regions in chromatin. We are using experimental data on chromatin modifications to find combinations of histone marks meaningful in predicting functional elements. A recent study we did together with Dr. Furlong group is a good example how Bayesian Network reconstruction can be used as a method for supervised classification of active enhancers in chromatin.
Genome-wide models of tissue-specific gene expression. For a long time now, we build and study genome wide models of gene expression. It started with our attempts to link gene expression clusters to sequence motifs. Later we have also looked at the relation between temporal patterns of transcription factor binding and gene expression during development. Most recently (in press), we have built a general probabilistic model for describing genome-wide patterns of tissue-specific gene expression allowing us to predict new genes activated at specific times or tissues in embryonic development.
Methods for Bayesian Network reconstruction. Together with Norbert Dojer, we are working on improving efficiency and accuracy of computational methods for Bayesian Network reconstruction from data. BNFinder, an efficient tool for Network topology reconstruction and classification is one of the tools we developed over the years.
Evolution of regulatory sequences. We are very interested in how did the regulatory domains arise in the course of evolution and how are they defined by DNA sequence elements. Our research covers both using information on conserved elements to find active regions, like in our Billboard method, as well as devising more general evolutionary models of regulatory sequences.
Modeling gene regulatory networks We want to understand how the same gene regulatory network in multiple cells gives rise to large-scale phenomena such as organ development. Earlier, we focused on stochastic models of gene regulatory networks, but our current interest shifted towards multi-cellular systems. Currently we don’t have models describing analytically both regulatory network at a molecular level and large scale dynamics of cell populations. However, with ever increasing computing power, we are able to simulate larger and larger populations with each cell equipped with simple stochastic regulatory network. Our tool, STOPS, can utilize modern computers (including GPU systems) to quickly simulate large cellular systems.
Analysis of genetic bases of Alzheimer’s disease. We are interested in identification of regulatory non-coding variants associated with Alzheimer’s disease and we explore the possibility of using feature selection and machine-learning classification algorithms for genotype-based classification of AD patients and healthy controls. More details can be found here.