ИСТИНА

Molecular modeling is important instrument in search of molecules, active against protein targets. Electrostatic interactions play crucial role both in the distant and near recognition of ligand by protein. Atomic charges - simple and convenient means to describe such interactions. There are a lot of methods for calculation of atomic charges. Among them empirical methods possess the following advantages: speed, interpretability, high precision in case good parametrization had been made. The main computational complexity of such methods is the solution of a system of linear equations (SLE), which size is proportional to atoms number (N) [1, 2]. Classic SLE solvers have complexity of O(N3). In this work, we investigate the application of iterative methods for computing empirical charges. Such methods have parallelization potential and less memory consumption due to option to use sparse matrices. We show that besides known advantages, use of iterative SLE solvers within DENR [2] approach makes it possible to achieve faster solution convergence due to special properties - diagonal predominance and sparseness - of in-method matrix [3]. The former one arises from the physical meaning of atomic electronegativity and hardness, whereas the last is caused by redistribution of charges along the molecule bonds. In result, at practical settings of the required precision, it allows us to effectively achieve quadratic computational complexity O(N2) against the cubic complexity O(N3) of classical methods. The latter opens the way for huge speedup and possibilities for fast online charges recalculations for the purposes of polarizable charges within molecular dynamics modeling.