Richard Wood - Evaluation of density functionals, SCC-DFTB, neglect of diatomic differential overlap (NDDO) models and molecular mechanics methods for prolyl-leucyl-glycinamide (PLG) and structural analogs

Version 1

      Publication Details (including relevant citation   information):

      Journal of Molecular Structure: THEOCHEM 944 (2010) 76–82   doi:10.1016/j.theochem.2009.12.026


      Prolyl-leucyl-glycinamide (PLG) is a unique endogenous peptide   that modulates dopamine receptor sub-types of the D2 receptor   family within the CNS. We seek to elucidate the structural basis   and molecular mechanism by which PLG and its analogs modulate   dopamine receptors, toward the development of new therapeutics to   treat Parkinson’s disease, tardive dyskinesia and schizophrenia.   As a first step toward establishing a validated protocol for   accurate computational modeling of PLG and associated   peptidomimetic analogs, we evaluated the accuracy of density   functional theory (DFT), wave function theory (WFT), and   molecular mechanics (MM) calculations for PLG and for a library   of structurally related small molecules. We first tested 12 local   and nonlocal density functionals, Hartree–Fock (HF) theory, four   ‘‘semiempirical” methods of the neglect of diatomic differential   overlap (NDDO) type, and one self-consistent-charge nonorthogonal   tight-binding (SCC-DFTB) method as implemented in two software   suites, against coupled-cluster benchmark geometries for   4-methylthiazolidine, a small molecule that comprises key   structural features present in our PLG-analog library. DFT and HF   calculations were done with the MG3S augmented polarized   triple-zeta basis set. We find that for 4-methylthiazolidine bond   distances,
      DFT significantly outperforms NDDO, and both SCC-DFTB versions we   evaluated perform worse than HF theory and are less accurate than   83% of the density functionals tested. The top five functionals   for 4-methylthiazolidine were M05-2X, mPW1PW, B97-2, M06-2X, and   PBEh, with mean unsigned errors (MUEs) in bond length of 0.0017,   0.0020, 0.0023, 0.0025 and 0.0027 Å, respectively. The widely   used B3LYP functional ranked 11th out of 12 functionals   evaluated, slightly below SCC-DFTB, and is significantly less   accurate for 4-methylthiazolidine bond distances (MUE = 0.0095 Å)   than the best local functional (M06-L, MUE = 0.0030 Å), which is   far less computationally costly. Based on that initial analysis,   we obtained new M05-2X benchmark geometric parameters for PLG and   a library of 11 peptidomimetic derivatives, which we in turn used   to examine the accuracy of thirty-four popular molecular   mechanics (MM) force fields, four NDDO approaches, and SCC-DFTB   for the full compound structures. Here, we found that ~70% of the   MM force fields tested superior to the best semiempirical and   SCC-DFTB codings. Moreover,
      AMBER-type force fields proved most accurate among MM methods for   this class of small-molecule peptidomimetics; the AMBER-type   methods comprised eight out of the top 10 molecular mechanics   options we tested.

      Address (URL): http://