Algorithms for macromolecular dynamics and constraint dynamics

TLDR

The protein trypsin inhibitor (BPTI), comprising 454 united atoms, serves as the model system. The study investigates applying molecular‑dynamics simulations to macromolecules, explores reducing computational effort via chain‑structure constraints, and proposes a predictor‑corrector scheme incorporating such constraints. Different integration algorithms are compared theoretically and practically, and a predictor‑corrector method incorporating bond constraints is introduced. Optimal algorithm choice depends on desired accuracy and force noise; for unconstrained dynamics the Gear predictor‑corrector performs best, while for constrained dynamics Gear and Verlet are comparable, bond‑length constraints cut computation time by a factor of three, and bond‑angle constraints are not recommended.

Abstract

Abstract The application of the computer simulation method of molecular dynamics to macromolecules is investigated. The protein trypsin inhibitor (BPTI), consisting of 454 united atoms, is used as an example. Different algorithms for integrating the equations of motion are compared, both theoretically and in practice. It is examined to what extent the chain structure of a macromolecule allows a reduction of the computational effort by the introduction of constraints in the dynamics of the chain. A calculational scheme is proposed, by which constraints can be incorporated in predictor-corrector algorithms. The optimum choice of an algorithm depends on the desired accuracy of the solution and on the character of the forces acting on the molecule, viz. whether these are noisy or not. For nonconstraint dynamics a Gear predictor-corrector algorithm yields the best results, whereas for constraint dynamics the Gear and Verlet algorithms produce comparable results. The application of bond-length constraints reduces the required computer time by a factor of 3. The inclusion of bondangle constraints is not recommended.