Protonate3D: assignment of ionization states and hydrogen coordinates to macromolecular structures.
Labute P. · 2009
Read Protonate3D: assignment of ionization states and hydrogen coordinates to macromolecular structures. by Labute P. (2009) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3056144.txt.
Introduction
Labute presented in 2009 a computer method that fills in the hydrogen atoms missing from most macromolecule structures. Experiments usually locate only the heavy atoms of proteins and bound drugs, yet hydrogen positions decide how molecules bind and react. His program, Protonate3D, places hydrogens automatically while also trying flipped side chains, alternate shapes, and different charge states for each group.
Its scoring weighs crowding, electric forces, solvent effects, and the cost of each charge state at a given acidity. Tests on known structures suggest the placements are accurate. Getting hydrogens right matters for drug design, since a flipped group or a wrong charge can change a predicted binding completely. The tool was built to prepare structures for the modeling work that follows structure solving. It handles ligands and solvent molecules alongside the main chain.
Catalog record
| CREATOR | Labute P. |
|---|---|
| YEAR | 2009 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 81 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Protonate3D: Assignment of ionization states and hydrogen coordinates to macromolecular structures By Paul Labute Published in: Proteins DOI: 10.1002/prot.22234 ABSTRACT A new method, called Protonate3D, is presented for the automated prediction of hydrogen coordinates given the 3D coordinates of the heavy atoms of a macromolecular structure. Protonate3D considers side-chain “flip,” rotamer, tautomer, and ionization states of all chemical groups, ligands, and solvent, provided suitable templates are available in a parameter file. The energy model includes van der Waals, Coulomb, solvation, rotamer, tautomer, and titration effects. The results of computational validation experiments suggest that Protonate3D can accurately predict the location of hydrogen atoms in macromolecular structures. Proteins 2009. © 2008 Wiley-Liss, Inc. INTRODUCTION As of October 2007, The protein data bank1 contained in excess of 45,000 structures, mostly the result of X-ray diffraction at resolution values greater than 1.5 Å. At these resolutions, the coordinates of hydrogen atoms cannot be observed and yet half of all atoms in the studied compounds are hydrogen atoms. Modern computational methodologies such as molecular mechanics, molecular dynamics, crystallographic refinement, electrostatic analysis, and docking require explicit (polar) hydrogen atoms for best results. Consequently, hydrogen atoms must be introduced into the X-ray structure data prior to undertaking these sorts of calculations. Thus, there is a need for automated procedures to predict hydrogen coordinates given the 3D coordinates of macromolecular structures. The prediction of hydrogen coordinates from hydrogen suppressed macromolecular structures, say proteins, is nontrivial: (a) rotamers of hydroxyls, phenols, thiols, methyls, and primary amines must be determined; (b) the ionization states of acidic and basic groups, such as carboxylic acids, amines, guanidines, imidazoles, and possibly phenols and even alcohols, must be determined; (c) the ionization states of transition metals must be assigned; (d) the orientation of water molecules must be determined; and (e) the tautomeric state of imidazoles and other moieties must be determined. Additionally, it is common to include the determination of terminal group “flips.” A flip is a conformational difference or element identity exchange, say, in the terminal amide of asparagine or glutamine. Because of limited resolution, the identities of oxygen and nitrogen often cannot be reliably determined; consequently, there may be an ambiguity in PDB crystal structures. Flip ambiguities can also exist for the imidazole rings in histidine, terminal sulfonamides, and phosphonamides. It is important to remember that protonation state prediction from static structures is motivated by the practical needs of molecular simulations or interpretation of structures. Inherently dynamic interactions of chemical groups with solvent (e.g., hydroxyl rotamers or weak acids and bases) cannot reasonably be expected to be predicted correctly in all cases. Instantaneous quantities, such as definite coordinates or ionization state, have little theoretical thermodynamic significance and, from a strict thermodynamic perspective, can be (at times) meaningless. In addition, coordinate errors in the nonhydrogen (heavy) atoms may result in unrealistic predictions; for example, unrealistically close crystallographic contacts exist in some structures. Notwithstanding these caveats, accurate prediction of protonation state and geometry has been the subject of much attention. In 2005, Forrest et al. compared a number of programs2 that predict hydrogen coordinates given heavy atom structures: MCCE,3 CHARMM,4 CNS,5 GROMACS,6 Reduce,7 WHAT IF,8 and X-PLOR.9 Subsequent to the publication of the study, the ICDA procedure was published.10 We will not repeat the results of the Forrest study here; however, we will make some broad methodological comparisons in the interests of placing this work into the context of prior methods. The protonation state/geometry prediction methods can be classified broadly according to whether (a) the state space search is systematic or stochastic, (b) the energy/scoring function is physicochemical or geometric/heuristic, and (c) there is a formal attempt at titration free energy optimization. GROMACS, CHARMM, and X-PLOR make no formal attempt to fully optimize the hydrogen interaction network. Discrete dihedral sampling and local force field-based interactions or a sequential, greedy buildup procedure is used in an effort to get a reasonable hydrogen placement. CNS places hydrogens at random and uses molecular dynamics and energy minimization to improve the configuration. MCCE uses a local systematic search followed by a Monte Carlo optimization procedure to place the hydrogens. A force field energy function is used for scoring with an optional solvation model (which dramatically increases the run time). The Reduce, WHAT IF, and ICDA programs employ a systematic search of torsion angles after first partitioning the system into clusters of atoms that are connected by potential hydrogen bonds. Reduce, WHAT IF, and ICDA depend critically on partitioning the system into small clusters each of which is subjected to a brute force optimization; for this reason, short interaction cutoffs on the order of hydrogen bond distances (3–4 Å) are required for tractable run times. Such short interaction cutoffs can be problematic for long-range ionic interactions, which fall off with the inverse of interionic distance; Reduce and WHAT IF use geometric/heuristic scoring and do not attempt formal titration and, consequently, this short cutoff assumption fits better with the approximate nature of the calculation. In contrast, ICDA uses an all-atom energy function with long-range Generalized Born electrostatics and yet a 3.1Å interaction cutoff appears to be used to partition the system into independent parts and long run times are suggested (20 min for 100 residues apparently on a 32 node cluster). Other than the functional form of the underlying energetic model, the central problem to solve is the optimization of the hydrogen/ion interaction network. Even if a continuous rotamer model is used, the tautomeric and ionization state space is discrete. This leads to a combinatorial optimization problem similar to that of macromolecular titration calculations11,12 in which the interactions of titratable groups affect the individual propensities for protonation. The particular optimization problem is similar to that of side-chain conformation prediction in proteins. The conformational preference for a particular side chain is influenced by interactions with other side chains whose conformation is unknown. Programs such as SCWRL13 use graph theory to reduce the complexity of the calculation. Dead-end elimination14 is often used to eliminate energetically poor states. Monte Carlo,15 group clustering methods and partitioning methods,10,11 and other methods16,17 have also been used to efficiently search the configurations for an optimal energy configuration. In this article, we present a new method—Protonate3D—which predicts hydrogen geometry, ionization, and tautomer states for macromolecular structures given the 3D coordinates of the nonhydrogen atoms. To our knowledge, Protonate3D is the first method that systematically optimizes the free energy of a macromolecular system in reasonable time, using a reasonably accurate electrostatic and implicit solvent model, and that takes longer range interactions into account, without partitioning the system into unrealistically small parts. We will describe the thermodynamic theory and a Unary Quadratic Optimization algorithm that obviates the need for unrealistically short interaction cutoffs and is the key to tractable run times. We will present the results of computational validation experiments and draw conclusions in the final section. THEORY AND METHODS Consider a macromolecular system of n (nonhydrogen) atoms with Cartesian coordinates. To rapidly evaluate the energy of a particular configuration of the system (including hydrogens), we will decompose the system into a collection of distinct chemical groups, {Ai}, consisting of atoms for which the protonation state is unknown and a set P, the part of the system for which there