Research

Based on the state-of-the-art of science, we assume that the nature of ligand binding to proteins depends significantly on the properties of the surrounding liquid medium, which is a full-fledged "player" in protein-ligand systems. Based on this assumption, we hypothesize that changes in the local structure of the liquid near the ligand (solute) in the liquid medium (solvent) depend on the properties of the solvent and lead to a change in the dynamic behavior of the components of the studied liquid-ligand systems. In the case of the solvent-ligand-protein system, the dissolution of the ligand in the protein pocket indirectly affects the properties of all components of the biofluid (water, saline, etc.) - ligand - protein system, accompanied by a reorganization of the local structure and dynamics of the liquid in the protein pocket. To test the above hypothesis, we will use molecular dynamics (MD).

In recent years, molecular informatics has transformed from a niche discipline into a driving force of the research and development of functional small molecules such as drugs and agrochemicals. Advanced algorithms as well as powerful computer hardware are now opening unprecedented opportunities for the targeted design of safe and efficacious small molecules. However, the full potential of computational methods in the biosciences is by far not exploited yet. One of the main reasons for this situation is the fact that the most powerful technologies in molecular informatics, machine learning and simulations in particular, depend on the availability of substantial amounts of high-quality data for development and validation. Despite recently launched initiatives to boost collaborative research and learning, the vast majority of high-quality chemical, biological and structural data remain behind corporate firewalls, inaccessible for research by experts in academia. This initiative for the Christian Doppler Laboratory for Molecular Informatics in the Biosciences seeks to push the frontiers of machine learning and molecular dynamics simulations technologies for the prediction of small-molecule bioactivity by supporting three expert academic research groups of the University of Vienna and the University of Natural Resources and Life Sciences (BOKU) with big data on the chemical and biological properties of small molecules, and with significant capacities for experimental testing and method validation. The unique synergy that will be generated by this consortium stems from two important factors: First, the two industry partners of this consortium have strong interest in cheminformatics but their business areas are non-competing. Second, and from a scientific point highly important, these industry partners focus on distinct chemical spaces, opening a unique opportunity for academics to boost the capacity and applicability of in silico methods with uniquely diverse, high-quality data.

Metal containing biomolecules are surprisingly common and essential for a spectrum of biological activities and physiological functions including i.a. respiration or photosynthesis. About one third of all the proteins include a metal-site, those metalloproteins typically coordinate metals by amino acid residues or organic co-factors. Metalloproteins have been investigated extensively towards understanding of their structure, function and, in particular, metal-ligand interactions which are relevant for drug design of metalloenzyme inhibitors and metallodrugs. Modelling and simulation of metalloproteins is challenging in various respects. Molecular dynamics (MD) simulations together with classical force fields do not suffice to describe the behaviour of metals and coordinated atoms. A quantum mechanical (QM) description of the systems is required to capture electronic effects. However, the efficiency of those methods is rather poor in the context of QM/MM hybrid approaches that are necessary to study large and complex biomolecules. To accelerate such hybrid systems, machine learning approaches seem to be promising. With the advances of deep learning algorithms, QM potential energy surfaces can be reproduced. Novel approaches in computational chemistry utilize neural networks (NNs) for the quantum description. With this project we propose a hybrid NN/MM-MD workflow, which we will implement in the GROMOS simulation package and apply the developed methodology to metal-sites of increasing complexity. Thus, we hope to improve the description of metal-ligand interactions in classical simulations with a specific focus on metalloproteins. The project opens the way for numerous applications and will allow for the evaluation of free-energy differences at a QM/MM level of theory, without the methodological challenges and computational costs. We expect that successful completion of the work will have considerable impact in the field of molecular simulations of metalloproteins.