Realistic sampling of amino acid geometries for a multipolar polarizable force field.

摘要: The Quantum Chemical Topological Force Field (QCTFF) uses the machine learning method kriging to map atomic multipole moments coordinates of all atoms in molecular system. It is important that operates on relevant and realistic training sets geometries. Therefore, we sampled single amino acid geometries directly from protein crystal structures stored Protein Databank (PDB). This sampling enhances conformational realism (in terms dihedral angles) However, these can be fraught with inaccurate bond lengths valence angles due artefacts refinement process X‐ray diffraction patterns, combined experimentally invisible hydrogen atoms. why developed a hybrid PDB/nonstationary normal modes (NM) approach called PDB/NM. superior over standard NM sampling, which captures only optimized stationary points acids gas phase. Indeed, PDB/NM combines chemically correct local Geometries using were used build models for alanine lysine, their prediction accuracy was compared built three other approaches. Bond length variation, as opposed variation angles, puts pressure accuracy, potentially lowering it. Hence, larger coverage does not deteriorate predictive models, around energetic minima so far development QCTFF. © 2015 Authors. Journal Computational Chemistry Published by Wiley Periodicals, Inc.