2.3. Method development for enhanced biological interpretation of gene expression data.

Responsible: Jörg Rahnenführer, MPI, Saarbrucken.

Background:

Increasing evidence suggests that only local structures of biological networks can be recovered from gene expression data. Taking known structural, regulatory or enzymatic roles of the corresponding proteins into consideration can improve the functional interpretation of the results significantly. Our methods can directly be extended and applied to other genomic high-throughput data, e.g. microarray-based CGH, a technique that will become increasingly important during NGFN2, with applications in cancer and medical genetics. Bayesian networks are graphical representations of the conditional independence structure among a set of variables, which are fitted to measured input-output behavior of a biological system, e.g. a part of a biochemical network model. Such measurements will be produced within NGFN2 by knock-down experiments applying RNAi technology, combined application of genomic (cytogenetic, CGH) and transcriptomic or proteomic investigations, epigenetic profiling and other methods. We aim to reconstruct small local networks by using Bayesian network inference based on such molecular data. Assessment of model fit and re-modeling will give further insights into cross-talk between pathways as well as subsystem properties.

Planned work:

Our primary objectives are the development and validation of methods that generate functional profiles with high biological or medical relevance. These profiles refer to gene sets with known biological meaning rather than to single genes. This provides fully interpretable snapshots of gene expression data obtained under specific conditions. A major application will be the characterization of disease types on a functional level. More specific goals are:

1. Identification of useful measures for co-regulation of genes in order to score whole sets of genes in microarray experiments and calculation of significance scores for such gene sets.
2. Development of validation procedures that are based on biological rather than purely statistical criteria.
3. Extension of scoring approach to discriminate between different phenotypes; development of scoring functions with discriminatory power.
4. Application of methods to other types of biological information like GO and MIPS annotations. The choice of meaningful gene sets will be done cooperation with SP 2.4 Gene Set Analysis.
5. Extension of the existing methods towards other multivariate high-throughput data given in matrix form, like CGH data, adapted to up-to-date demands within NGFN-2.
6. Modelling of local pathways, like parts of the EGF/MAP kinase pathway or tyrosine kinase dependent cell proliferation signals for inferring small parts of biochemical networks, inference of small netwoeks with Bayesian inference.
7. Conduction of case studies on a variety of real gene expression data from NGFN-2 members and evaluation of the biological findings.