Hamacher

Computational Biology and Simulation

Prof. Dr. Kay Hamacher

Research

Our ultimate goal is the development of an in silico molecular simulation and analysis pipeline for synthetic biology research and molecular systems biology to a) investigate biological and biophysical phenomena on multiple scales and b) to engineer biomolecules and functional biomaterials. To this end we focus on integrating our techniques for efficient simulations, coarse-grained models, and structure modeling approaches with off-the-shelf, sequence-based bioinformatics software packages. To investigate the effects of molecular properties on cell biology the laboratory also develops algorithms to efficiently simulate molecular networks on multi-core architectures.

Reduced Molecular Models (RMM) and Multiscale Simulations

We developed a new methodology that closes the gap between Normal Mode Analysis, elastic network models, and their statistical mechanism (in particular multivariate distributions). While the former approach provides for amino acid specificity the latter models are computational inexpensive and therefore suitable for large-scale screening efforts in structural-functional genomics, annotation of stochastic evolutionary dynamics, and systems biology. In a first application we were able to show effects of mutations towards drug resistance in HIV1 protease and the underlying mechanism.
In an extended model we were able to show the potential of RMMs for e.g. macromolecular assembly-processes. In particular we derived the assembly map of the bacterial ribosome based solely on structural properties. This method development effort includes atomistic molecular dynamics and quantum chemical calculations for parameterization and cross-validation.

Biological Application of RMM: Functional Annotation of Molecular Phenotypes & Design of Functional Molecules

With the RMM approach we were able to annotate the molecular evolution of the HIV1 protease towards drug resistance from a functional perspective. The integration of sequence information with the mechanics/dynamics of the protein is conceptually promising, as the pure sequence information is concerned with solely the mutational event, while the selective advantage of such a mutation can only be investigated in the physical realm/the molecular phenotype.
Besides this functional annotation we proposed some potential targets to reduce the enzymatic function of the protease, while at the same time reducing the propensity of the viral enzyme to develop drug resistance. The correctness of this theoretical finding was independently shown in vitro recently.
Currently work on other protein families and their evolution is under way, e.g. in collaboration with the Thiel laboratory. Here the focus lies, however, also on the design of functional, molecular biosensors. Our RMM results are in good agreement with functional, in vivo data obtained in the same lab. Additionally sequence based bioinformatics is used to understand structure and (co)evolution of genomic and membrane protein sequences of algae and their viruses.
We also are engaged in understanding the evolution and function of gas vesicles. Work is underway to predict protein structures of gas vesicle proteins, as well as the gas vesicle formation mechanism at the molecular level. This work is done in collaboration with the Pfeifer lab and researchers from outside the TUD. The understanding of such mesoscopic assemblies will afterwards be possible by our RMM approach.
The mode of action of functional ligands can conceptually be separated into two distinct classes: a) binding to active sites, and thus preventing further e.g. enzymatic action or b) allosteric effects that change the mechanics/dynamics of the parti-cipating molecules. Combining a previous study from our lab on information theo-retical inspired ligand design with RMMs this procedure is capable of consistently classifying the allosteric action of various ligand topologies. The method is computationally efficient and avoids several problems usually encountered when employing e.g. time series analysis of molecular trajectories.

Molecular Network Evolution & Design

In collaboration with the Goesele group (Dept. of CS) we implemented a massively parallel simulation approach on GPUs to investigate the (oscillatory) dynamics, (bi-)stability, robustness, and evolvability of molecular networks. Scanning high-dimensional parameter spaces becomes thus feasible on standard PC clusters. In combination with optimization approaches (s. below) we are now able to design biochemical and regulatory networks for particular purposes in an engineering fashion. The practical implications of such designed and of naturally occurring, evolved networks haven been investigated e.g. in a joint project with Honda Research.

Stochastic Global Optimization

The determination of the global optimum of a free energy function is one of several approaches to protein structure prediction. Optimization is also used in molecular docking, molecular modeling, and ligand design. We developed an algorithm that proved to be some two orders of ma-gnitude more efficient in chemical/physical applications than Kirckpatrick's widely used simulated annealing method. A recently developed adaptive version show-ed an increased efficiency of another order of magnitude. This meta-algorithm turned out to be transferable to other optimization techniques such as Extremal Optimization. This increased efficiency will allow us to reach several intermediate steps of our ultimate goal of an integrated pipeline: efficient optimization procedures are key to structure modeling, while the design of networks relies heavily on optimization techniques.