Michael Schroeder

Michael Schroeder, Biotec, TU-Dresden Networks play a crucial role in computational biology. We are interested in developing algorithms to analyse large-scale protein interaction and regulatory networks including functional annotation with text-mining, ontologies, and reasoning. Together with experimental and clinical groups, we apply these techniques to interpret disease networks in cancer and neurodegeneration.

We purse three strands of works:

1. Power Graphs: Power graphs identify motifs such as cliques and bi-cliques in networks such as protein interaction networks, regulatory networks, genetic networks, protein-ligand interaction networks, and homology networks. They reduce the complexity of these networks by up to 90 percent and allow to identify complexes, shared domains, and binding motifs.

2. Structural protein interaction networks: We developed SCOPPI.org, the a database for structural protein interaction interfaces. We systematically characterized protein binding sites and established that 40% of domain families bind in multiple orientations, that binding orientation is not conserved in one third of gene fusions, that ancient interfaces are enriched by symmetric homodimers. We also developed an algorithm to identify convergently evolved interfaces, which allows us to identify many examples of viruses mimicking native interfaces. Using structural interactions, we predicted an interaction between a serine protease up-regulated in pancreas cancer and linked to formation of metastases and an inhibitor, which is down-regulated in pancreas cancer.

3. Functional annotation with text-mining, ontologies, and reasoning: We developed GoPubMed.org, an ontology-based literature search engine for functional characterisation, which led to the spin-off of Transinsight GmbH. The underlying algorithms to identify biolgical entities in text are world-wide leading as shown in the BioCreative 2 text-mining challenge.