Přehled o publikaci

Current advances in hardware and algorithm develop- ment allow the life science researchers to replace the experiment with a computer simulation. A key ob- ject of these computations is a molecule - a group of atoms interconnected via a cloud of electrons. For its computational processing, electrons around the atom are often represented by one number: partial atomic charge. It can be calculated by quantum mechan- ics (QM), which offers high accuracy at the cost of long computation time, or markedly faster by empirical methods such as Electronegativity Equalization Method (EEM). Empirical methods calibrate their parameters to the particular QM charge calculation approach by multi-dimensional optimization procedure. This work systematically summarizes and compares the accuracy and computational performance of available EEM pa- rameterization approaches with local, global or com- bined optimization (least squares, evolutionary and ge- netic algorithms). Moreover, we propose a new method- ology called guided minimization. We found that local optimization plays a crucial role in the parametrization, and only methodologies combining a global and a lo- cal optimization provide high-quality EEM parameters. Furthermore, we observed that global iterations of evo- lutionary of genetic algorithm often do not contribute to the result. Therefore, we reduced the global search method to guided minimization that achieves same or better accuracy than state-of-the-art methods and sur- passes them with simplicity and speed.