Research

Parallel numerical algorithms and simulations

Parallel numerical algorithms and simulations

V dnešní době je mnoho problémů z reálného světa řešeno pomocí simulací. Díky simulacím je možné například lépe chápat komplexní interakce biomolekul, trénovat operace ve virtuálním prostředí či odhadovat dosah přírodních katastrof. Mnohé simulace jsou však i pro současné počítače extrémně náročné, je proto třeba hledat vhodné matematické modely a paralelizovat, akcelerovat a optimalizovat jejich implementaci. V laboratoři se zabýváme především metodami pro urychlování kódu a jejich automatizací, výpočetní chemií a medicínskými simulacemi.
 
 


Methods and algorithms for direct ligand annotation

Biomacromolecules (e.g. proteins, nucleic acids) are large molecules of biological origins that constitute various parts of living organisms. They carry out their functions with the aid of ligands – small molecules, which origin is biological as well. To facilitate the research of biomacromolecules and their complexes with ligands, it is necessary to obtain a model of their real structure. The amount of structures that are being submitted to freely available databases is steadily increasing. However, with the rise of the count of newly assembled structure models, the amount of errors in them is rising as well. The scientific community’s reaction to this newly emerged problem was the development of validation software. Students and alumni of the laboratory took major part in developing MotiveValidator (published here) – a software tool for ligand validation. They also created ValidatorDB (published here) – a publicly accessible database of validation results of ligands that are stored in the Protein Data Bank database (the biggest database of its kind). This validation of ligands showed that approximately 17 % of ligands is missing group(s) of atoms, has redundant group(s) of atoms, or has incorrect spatial configuration of atoms adjacent to chiral center(s) (source).

A major problem of ligand quality is their incorrect annotation. Nontrivial percentage of ligands in databases is annotated in a way that does not correspond to their deposited structure. To tackle this problem, we are developing effective algorithms for direct ligand annotation. Unlike several solutions that already exist, this approach does not require a reference set of ligands. That is a key advantage since the quality of reference ligands sometimes leaves a lot to be desired.

Another issue that has to be addressed while directly annotating ligands is the quality of their actual structure. Therefore, an extensive analysis that explores relationships between diverse structure factors has been performed first. We were trying to ascertain whether the quality of ligands is on the rise as much as is the quality of biomacromolecules. Furthermore, we were trying to determine factors that influence the structure quality the most.

Contact: Vladimír Horský, vladimir.horsky@mail.muni.cz


Computation of Atomic Charges

Partial atomic charges describe the distribution of electron density in a molecule, and therefore they provide clues regarding the chemical behavior of molecules. Atomic charges are frequently used in molecular modeling applications and they have also become popular chemoinformatics descriptors. Because of not corresponding to any physical quantity directly, the partial atomic charges cannot be measured experimentally. However, they can be computed. Standard way of their calculation involves quantum mechanics (QM), methods which make no further assumptions and which use only physical constants. The principal disadvantage of QM methods is their computational complexity which limits their practical use to small systems only.

Markedly faster alternative to the QM approaches is represented by the empirical methods. However, improved time complexity comes at a price – empirical methods utilize empirical parameters which often have to be obtained from QM charges in a process called parameterization. Unfortunately, there is no universally applicable charge calculation method or even a parameterization scheme – for some applications the most suitable charges are derived using one method, for other applications by using another one. Therefore, we focus on study, design and implementation of empirical charge calculation methods and their parameterizations.

Contact: Tomáš Raček, tom@krab1k.net


Molecular Simulations with Adaptive Potentials

Computational chemistry allows researchers to experiment in sillico: by running computer simulations of biological or chemical processes of interest. However, the validity of simulation results depends on used potentials: equations, functions and their parameters that together approximate the given process. The values of parameters usually stem from quantuum mechanics computations and a constant value is used for molecular simulation. Many simulations will benefit and increase their validity if we can adjust the parameters‘ values as the simulation is running and the structure of the simulated system is changing. The adaptation of parameters introduces new degrees of freedom into the model and presents a computational challenge from the point of numerical stability.


trpcage

The research at this topic is conducted in cooperation with Radka Svobodová Vařeková (Faculty of Science, Masaryk University, Brno) and Vojtěch Spiwok (Department of Biochemistry and Microbiology, Intitute of Chemical technology, Prague).

Contact: pazurikova@ics.muni.cz

(c) 2021 SITOLA, Administration