Development of an automated fragment molecular orbital (FMO) calculation protocol toward construction of quantum mechanical calculation database for large biomolecules

Chiduru Watanabe, Hirofumi Watanabe, Yoshio Okiyama, Daisuke Takaya, Kaori Fukuzawa, Shigenori Tanaka, Teruki Honma
2019 Chem-Bio Informatics Journal  
We developed an automated FMO calculation protocol (Auto-FMO protocol) to calculate huge numbers of protein and ligand complexes, such as drug discovery targets, by an ab initio FMO method. The protocol performs not only FMO calculations but also pre-processing of input structures by homology modeling of missing atoms and subsequent MM-based optimization, as well as post-processing of calculation results. In addition, QM/MM optimization of complex structures, conformational searches of ligand
more » ... earches of ligand structures in solvent, and MM-PBSA/GBSA calculations can be optionally carried out. In this paper, FMO calculations for 149 X-ray complex structures of estrogen receptor α and p38 MAP kinase were performed at the K computer and in-house PC cluster server by using the Auto-FMO protocol. To demonstrate the usefulness of the Auto-FMO protocol, we compared the ligand binding interaction energies by the Auto-FMO protocol with those of manually prepared data. In most cases, the data calculated by the Auto-FMO protocol showed reasonable agreement with the manually prepared data. Further improvement of the protocol is necessary for the treatment of ionization and tautomerization at the structure preparation stage, because some outlier data were observed due to these issues. The Auto-FMO protocol provides a powerful tool to deal with huge numbers of complexes for drug design, as well as for the construction of the FMO database ( released in 2019. 6 Key Words: Fragment molecular orbital (FMO), intermolecular interaction, ligand binding energy, estrogen receptor α (ERα), p38 mitogen-activated protein (MAP) kinase, FMO database Area of Interest: Molecular recognition and molecular modeling Introduction Ab initio quantum mechanical calculations for whole large biomolecules can be efficiently performed by the fragment molecular orbital (FMO) method [1] [2] [3] . An inter-fragment interaction energy (IFIE) analysis based on FMO calculations can easily represent the detailed interactions in fragment units. The FMO method is already recognized as a useful drug design tool to analyze ligand binding interactions, incorporating electrostatic interactions such as hydrogen bonds and dispersion forces such as CH/π interactions, using the pair interaction energy decomposition analysis (PIEDA) [4, 5] and fine fragmentation by the functional group unit, rather than the amino acid residue unit and the whole ligand [6] [7] [8] . Recently, the IFIE analysis and its energy decomposition analysis have been applied to the prediction of binding affinity for rational drug design [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] . Using FMO calculations of tens of complexes for one target protein, the essential and characteristic interactions of the ligand binding mode can be abstracted from the IFIE and PIEDA data by clustering methods [19] [20] and singular value decomposition [21] . In addition, the prediction of the activity cliff, which is very difficult using conventional molecular mechanics (MM)-based scoring functions, such as Glide score and molecular mechanics Poisson-Boltzmann surface area (MM-PBSA), was successfully accomplished by the FMO method with molecular mechanics Poisson-Boltzmann surface area (FMO+MM-PBSA) approach incorporating MM-based desolvation effects, using protein-ligand complexes optimized by the quantum mechanics/molecular mechanics (QM/MM) method [22] . Moreover, the FMO based polarizable continuum model (FMO-PCM) [23] or FMO based Poisson-Boltzmann surface area (FMO-PBSA) [24] [25] [26] methods provide more reliable results in solution, by using a fully-polarizable medium for the solute. Since 2014, we have performed FMO calculations for various drug discovery targets, such as kinases, nuclear receptors, proteases, and protein-protein interactions (PPIs), with experimental binding affinities (IC 50 , K i , K d values) as the activities of the FMO drug design consortium (FMODD) [27] . To calculate the huge number of different structures by a manual procedure, we must investigate and choose the various modeling conditions and FMO settings one by one. For example, appropriate structure preparation, which includes complementation of missing atoms or missing residues, addition of hydrogen atoms, and structure minimization, is critically important as the preprocessing before the FMO calculations. However, appropriate methods for preprocessing have not yet been established. We discussed the modeling conditions for the complementation of heavy atoms, with/without water molecules, and the restraint of heavy atoms on the minimization in the FMODD consortium. As a result, some modeling case studies have been reported [28] [29] [30] [31] [32] [33] [34] . Another issue is the treatment of a large amount of structure data, including more than 145,000 protein data bank (PDB) entries, to construct an FMO database [35] in the future. There are limits to human power in preparing a huge number of structures by a manual operation. In addition, it is not easy to appropriately perform FMO calculations for inexperienced researchers, in terms of the structure preparation and FMO settings. Thus, we have started to develop "an automated FMO calculation protocol" (Auto-FMO protocol). In this paper, we constructed the Auto-FMO protocol consisting of structural preparation based on the MM method, structural optimization based on the QM/MM calculations with our own N-layered integrated molecular orbital and molecular
doi:10.1273/cbij.19.5 fatcat:eqi5qtlb4ngjbfg6ox5zgxacla