Use of Extended-Hückel Descriptors for Rapid and Accurate Predictions of Conjugated Torsional Energy Barriers.

Over the past few decades, virtual high-throughput screening (vHTS) and molecular dynamics simulations have become effective and widely used tools in the initial stages of drug discovery efforts. These methods allow a great number of druglike molecules to be screened quickly and inexpensively. Unfortunately, however, the accuracies of both these methods rely on the quality of the underlying molecular mechanics force fields (FFs), which are often poor. This major weakness originates from the reliance of FFs on a finite list of specific parameters, called atom types, which have low transferability between molecules. In particular, the torsional energy barriers of druglike molecules are notoriously difficult to predict. Continuing our endeavor to understand factors affecting the torsional energy barriers of small molecules and quantify them, we showed that descriptors calculated using the extended-Hückel method could be used to rapidly assign accurate torsion parameters for conjugated molecules. This method, called H-TEQ 4.5, was developed using a set of 684 conjugated molecules. It was subsequently validated on a test set of 200 diverse molecules and produced an average root-mean-square error (rmse) of 1.01 kcal·mol-1, with respect to the reference quantum mechanic torsional profiles. For comparison, GAFF2, MMFF94, and MAB produced average rmse's of 3.49, 1.50, and 1.77 kcal·mol-1, respectively. H-TEQ 4.5 is also computationally inexpensive, running just under 0.25 ms for a biphenyl molecule on a home computer, allowing it to be used for vHTS of large libraries of compounds. Overall, H-TEQ 4.5 solved the problems associated with the transferability of torsion parameters for conjugated molecules. This method was incorporated into the Molecular Operating Environment and will be available for a wide variety of applications.

Atom Type Independent Modeling of the Conformational Energy of Benzylic, Allylic, and Other Bonds Adjacent to Conjugated Systems.

Applications of computational methods to predict binding affinities for protein/drug complexes are routinely used in structure-based drug discovery. Applications of these methods often rely on empirical force fields (FFs) and their associated parameter sets and atom types. However, it is widely accepted that FFs cannot accurately cover the entire chemical space of drug-like molecules, due to the restrictive cost of parametrization and the poor transferability of existing parameters. To address these limitations, initiatives have been carried out to develop more transferable methods, in order to allow for more rigorous descriptions of any drug-like molecule. We have previously reported H-TEQ, a method which does not rely on atom types and incorporates well established chemical principles to assign parameters to organic molecules. The previous implementation of H-TEQ (a torsional barrier prediction method) only covered saturated and lone pair containing molecules; here, we report our efforts to incorporate conjugated systems into our model. The next step was the evaluation of the introduction of unsaturations. The developed model (H-TEQ3.0) has been validated on a wide variety of molecules containing heteroaromatic groups, alkyls, and fused ring systems. Our method performs on par with one of the most commonly used FFs (GAFF2), without relying on atom types or any prior parametrization.

Torsional Energy Barriers of Biaryls Could Be Predicted by Electron Richness/Deficiency of Aromatic Rings; Advancement of Molecular Mechanics toward Atom-Type Independence.

Biaryl molecules are ubiquitous pharmacophores found in natural products and pharmaceuticals. In spite of this, existing molecular mechanics force fields are unable to accurately reproduce their torsional energy profiles, except for a few well-parametrized cases. This effectively limits the ability of structure-based drug design methods to correctly identify hits involving biaryls with confidence (e.g., during virtual screening, employing docking and/or molecular dynamics simulations). Continuing in our endeavor to quantify organic chemistry principles, we showed that the torsional energy profile of biaryl compounds could be computed on-the-fly based on the electron richness/deficiency of the aromatic rings. This method, called H-TEQ 4.0, was developed using a set of 131 biaryls. It was subsequently validated on a separate set of 100 diverse biaryls, including multisubstituted, bicyclic and tricyclic druglike molecules, and produced an average root-mean-square error (RMSE) of 0.95 kcal·mol-1. For comparison, GAFF2 produced an RMSE of 3.88 kcal·mol-1, owing to problems associated with the transferability of torsion parameters. The success of H-TEQ 4.0 provided further evidence that force fields could transition to become atom-type independent, providing that the correct chemical principles are used. Overall, this method solved the problem of transferability of biaryl torsion parameters, while simultaneously improving the overall accuracy of the force field.

Atom Types Independent Molecular Mechanics Method for Predicting the Conformational Energy of Small Molecules.

We previously implemented a well-known qualitative chemical principle into an accurate quantitative model computing relative potential energies of conformers. According to this principle, hyperconjugation strength correlates with electronegativity of donors and acceptors. While this earlier version of our model applies to σ bonds, lone pairs, disregarded in this earlier version, also have a major impact on the conformational preferences of molecules. Among the well-established principles used by organic chemists to rationalize some organic chemical behaviors are the anomeric effect, the alpha effect, basicity, and nucleophilicity. These effects are directly related to the presence of lone pairs. We report herein our effort to incorporate lone pairs into our model to extend its applicability domain to any saturated small molecules. The developed model H-TEQ 2 has been validated on a wide variety of molecules from polyaromatic molecules to carbohydrates and molecules with high heteroatoms/carbon ratios. Interestingly, this method, in contrast to common force field-based methods, does not rely on atom types and is virtually applicable to any organic molecules.

Elucidating Hyperconjugation from Electronegativity to Predict Drug Conformational Energy in a High Throughput Manner.

Computational chemists use structure-based drug design and molecular dynamics of drug/protein complexes which require an accurate description of the conformational space of drugs. Organic chemists use qualitative chemical principles such as the effect of electronegativity on hyperconjugation, the impact of steric clashes on stereochemical outcome of reactions, and the consequence of resonance on the shape of molecules to rationalize experimental observations. While computational chemists speak about electron densities and molecular orbitals, organic chemists speak about partial charges and localized molecular orbitals. Attempts to reconcile these two parallel approaches such as programs for natural bond orbitals and intrinsic atomic orbitals computing Lewis structures-like orbitals and reaction mechanism have appeared. In the past, we have shown that encoding and quantifying chemistry knowledge and qualitative principles can lead to predictive methods. In the same vein, we thought to understand the conformational behaviors of molecules and to encode this knowledge back into a molecular mechanics tool computing conformational potential energy and to develop an alternative to atom types and training of force fields on large sets of molecules. Herein, we describe a conceptually new approach to model torsion energies based on fundamental chemistry principles. To demonstrate our approach, torsional energy parameters were derived on-the-fly from atomic properties. When the torsional energy terms implemented in GAFF, Parm@Frosst, and MMFF94 were substituted by our method, the accuracy of these force fields to reproduce MP2-derived torsional energy profiles and their transferability to a variety of functional groups and drug fragments were overall improved. In addition, our method did not rely on atom types and consequently did not suffer from poor automated atom type assignments.

Understanding P450-mediated Bio-transformations into Epoxide and Phenolic Metabolites.

Adverse drug reactions are commonly the result of cytochrome P450 enzymes (CYPs) converting the drugs into reactive metabolites. Thus, information about the CYP bioactivation of drugs would not only provide insight into metabolic stability, but also into the potential toxicity. For example, oxidation of phenyl rings may lead to either toxic epoxides or safer phenols. Herein, we demonstrate that the potential to form reactive metabolites is encoded primarily in the properties of the molecule to be oxidized. While the enzyme positions the molecule inside the binding pocket (selects the site of metabolism), the subsequent reaction is only dependent on the substrate itself. To test this hypothesis, we used this observation as a predictor of drug inherent toxicity. This approach was used to successfully identify the formation of reactive metabolites in over 100 drug molecules. These results provide a new perspective on the impact of functional groups on aromatic oxidation of drugs and their effects on toxicity.

Assessment of fully automated antibody homology modeling protocols in molecular operating environment.

The success of antibody-based drugs has led to an increased demand for predictive computational tools to assist antibody engineering efforts surrounding the six hypervariable loop regions making up the antigen binding site. Accurate computational modeling of isolated protein loop regions can be quite difficult; consequently, modeling an antigen binding site that includes six loops is particularly challenging. In this work, we present a method for automatic modeling of the FV region of an immunoglobulin based upon the use of a precompiled antibody x-ray structure database, which serves as a source of framework and hypervariable region structural templates that are grafted together. We applied this method (on common desktop hardware) to the Second Antibody Modeling Assessment (AMA-II) target structures as well as an experimental specialized CDR-H3 loop modeling method. The results of the computational structure predictions will be presented and discussed.

AMBER Free Energy Tools: A New Framework for the Design of Optimized Alchemical Transformation Pathways.

We develop a framework for the design of optimized alchemical transformation pathways in free energy simulations using nonlinear mixing and a new functional form for so-called "softcore" potentials. We describe the implementation and testing of this framework in the GPU-accelerated AMBER software suite. The new optimized alchemical transformation pathways integrate a number of important features, including (1) the use of smoothstep functions to stabilize behavior near the transformation end points, (2) consistent power scaling of Coulomb and Lennard-Jones (LJ) interactions with unitless control parameters to maintain balance of electrostatic attractions and exchange repulsions, (3) pairwise form based on the LJ contact radius for the effective interaction distance with separation-shifted scaling, and (4) rigorous smoothing of the potential at the nonbonded cutoff boundary. The new softcore potential form is combined with smoothly transforming nonlinear λ weights for mixing specific potential energy terms, along with flexible λ-scheduling features, to enable robust and stable alchemical transformation pathways. The resulting pathways are demonstrated and tested, and shown to be superior to the traditional methods in terms of numerical stability and minimal variance of the free energy estimates for all cases considered. The framework presented here can be used to design new alchemical enhanced sampling methods, and leveraged in robust free energy workflows for large ligand data sets.

Variability in docking success rates due to dataset preparation.

The results of cognate docking with the prepared Astex dataset provided by the organizers of the "Docking and Scoring: A Review of Docking Programs" session at the 241st ACS national meeting are presented. The MOE software with the newly developed GBVI/WSA dG scoring function is used throughout the study. For 80 % of the Astex targets, the MOE docker produces a top-scoring pose within 2 Å of the X-ray structure. For 91 % of the targets a pose within 2 Å of the X-ray structure is produced in the top 30 poses. Docking failures, defined as cases where the top scoring pose is greater than 2 Å from the experimental structure, are shown to be largely due to the absence of bound waters in the source dataset, highlighting the need to include these and other crucial information in future standardized sets. Docking success is shown to depend heavily on data preparation. A "dataset preparation" error of 0.5 kcal/mol is shown to cause fluctuations of over 20 % in docking success rates.

Antibody modeling assessment.

A blinded study to assess the state of the art in three-dimensional structure modeling of the variable region (Fv) of antibodies was conducted. Nine unpublished high-resolution x-ray Fab crystal structures covering a wide range of antigen-binding site conformations were used as benchmark to compare Fv models generated by four structure prediction methodologies. The methodologies included two homology modeling strategies independently developed by CCG (Chemical Computer Group) and Accerlys Inc, and two fully automated antibody modeling servers: PIGS (Prediction of ImmunoGlobulin Structure), based on the canonical structure model, and Rosetta Antibody Modeling, based on homology modeling and Rosetta structure prediction methodology. The benchmark structure sequences were submitted to Accelrys and CCG and a set of models for each of the nine antibody structures were generated. PIGS and Rosetta models were obtained using the default parameters of the servers. In most cases, we found good agreement between the models and x-ray structures. The average rmsd (root mean square deviation) values calculated over the backbone atoms between the models and structures were fairly consistent, around 1.2 Å. Average rmsd values of the framework and hypervariable loops with canonical structures (L1, L2, L3, H1, and H2) were close to 1.0 Å. H3 prediction yielded rmsd values around 3.0 Å for most of the models. Quality assessment of the models and the relative strengths and weaknesses of the methods are discussed. We hope this initiative will serve as a model of scientific partnership and look forward to future antibody modeling assessments.

Structure-based charge calculations for predicting isoelectric point, viscosity, clearance, and profiling antibody therapeutics.

The effect of hydrophobicity on antibody aggregation is well understood, and it has been shown that charge calculations can be useful for high-concentration viscosity and pharmacokinetic (PK) clearance predictions. In this work, structure-based charge descriptors are evaluated for their predictive performance on recently published antibody pI, viscosity, and clearance data. From this, we devised four rules for therapeutic antibody profiling which address developability issues arising from hydrophobicity and charged-based solution behavior, PK, and the ability to enrich for those that are approved by the U.S. Food and Drug Administration. Differences in strategy for optimizing the solution behavior of human IgG1 antibodies versus the IgG2 and IgG4 isotypes and the impact of pH alterations in formulation are discussed.

LowModeMD--implicit low-mode velocity filtering applied to conformational search of macrocycles and protein loops.

We present a method for conformational search of complex molecular systems such as macrocycles and protein loops. The method is based on perturbing an existing conformation along a molecular dynamics trajectory using initial atomic velocities with kinetic energy concentrated on the low-frequency vibrational modes, followed by energy minimization. A novel Chebyshev polynomial filter is used to heavily dampen the high-frequency components of a randomly generated Maxwell-Boltzmann velocity vector. The method is very efficient, even for large systems; it is straightforward to implement and requires only standard force-field energy and gradient evaluations. The results of several computational experiments suggest that the method is capable of efficiently sampling low-strain energy conformations of complex systems with nontrivial nonbonded interaction networks.

Training a scoring function for the alignment of small molecules.

A comprehensive data set of aligned ligands with highly similar binding pockets from the Protein Data Bank has been built. Based on this data set, a scoring function for recognizing good alignment poses for small molecules has been developed. This function is based on atoms and hydrogen-bond projected features. The concept is simply that atoms and features of a similar type (hydrogen-bond acceptors/donors and hydrophobic) tend to occupy the same space in a binding pocket and atoms of incompatible types often tend to avoid the same space. Comparison with some recently published results of small molecule alignments shows that the current scoring function can lead to performance better than those of several existing methods.

Pocket similarity: are alpha carbons enough?

A novel method for measuring protein pocket similarity was devised, using only the alpha carbon positions of the pocket residues. Pockets were compared pairwise using an exhaustive three-dimensional Calpha common subset search, grouping residues by physicochemical properties. At least five Calpha matches were required for each hit, and distances between corresponding points were fit to an Extreme Value Distribution resulting in a probabilistic score or likelihood for any given superposition. A set of 85 structures from 13 diverse protein families was clustered based on binding sites alone, using this score. It was also successfully used to cluster 25 kinases into a number of subfamilies. Using a test kinase query to retrieve other kinase pockets, it was found that a specificity of 99.2% and sensitivity of 97.5% could be achieved using an appropriate cutoff score. The search itself took from 2 to 10 min on a single 3.4 GHz CPU to search the entire Protein Data Bank (133 800 pockets), depending on the number of hits returned.

Protonate3D: assignment of ionization states and hydrogen coordinates to macromolecular structures.

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.

The generalized Born/volume integral implicit solvent model: estimation of the free energy of hydration using London dispersion instead of atomic surface area.

A new generalized Born model for estimating the free energy of hydration is presented. The new generalized Born/volume integral (GB/VI) estimates the free energy of hydration as a classical electrostatic energy plus a cavitation energy that is not based upon atomic surface area (SA) used in GB/SA hydration models but on a VI London dispersion energy estimated from quantities already calculated in the classical electrostatic energy. The (relatively few) GB/VI model parameters are fitted to experimental data, and parameterizations for two different atomic partial charge models are presented. Comparison of the calculated and experimental free energies of hydration for 560 small molecules (both neutral and charged) shows good agreement (r(2) = 0.94).

Methods of Exploring Protein-Ligand Interactions to Guide Medicinal Chemistry Efforts.

We present a number of techniques to analyze protein-ligand interactions in the context of medicinal chemistry: crystal Contract Preferences, Electrostatic Maps and pharmacophore screening using Hückel Theory. Contact Preferences is a statistical technique to predict hydrophobic and hydrophilic geometry in receptor active sites. Electrostatic Maps use the Poisson-Boltzmann Equation to model solvation effects and are particularly useful for predicting hydrophobic regions. Pharmacophore annotation with Hückel Theory provides finer detail of hydrogen bonding interactions, including CH..O interactions. Applications to AblK:Gleevec and CDK2 virtual screening are presented.

A probabilistic approach to high throughput drug discovery.

A methodology is presented in which high throughput screening experimental data are used to construct a probabilistic QSAR model which is subsequently used to select building blocks for a virtual combinatorial library. The methodology is based upon statistical probability estimation and not regression. The methodology is applied to the construction of two focused virtual combinatorial libraries: one for cyclic GMP phosphodiesterase type V inhibitors and one for acyl-CoA:cholesterol O-acyltransferase inhibitors. The results suggest that the methodology is capable of selecting combinatorial substituents that lead to active compounds starting with binary (pass/fail) activity measurements.

Derivation and applications of molecular descriptors based on approximate surface area.

Three sets of molecular descriptors that can be computed from a molecular connection table are defined. The descriptors are based on the subdivision and classification of the molecular surface area according to atomic properties (such as contribution to logP, molar refractivity, and partial charge). The resulting 32 descriptors are shown (a) to be weakly correlated with each other; (b) to encode many traditional molecular descriptors; and (c) to be useful for QSAR, QSPAR, and compound classification.

A Cavity Corrected 3D-RISM Functional for Accurate Solvation Free Energies.

We show that an Ng bridge function modified version of the three-dimensional reference interaction site model (3D-RISM-NgB) solvation free energy method can accurately predict the hydration free energy (HFE) of a set of 504 organic molecules. To achieve this, a single unique constant parameter was adjusted to the computed HFE of single atom Lennard-Jones solutes. It is shown that 3D-RISM is relatively accurate at predicting the electrostatic component of the HFE without correction but requires a modification of the nonpolar contribution that originates in the formation of the cavity created by the solute in water. We use a free energy functional with the Ng scaling of the direct correlation function [Ng, K. C. J. Chem. Phys.1974, 61, 2680]. This produces a rapid, reliable small molecule HFE calculation for applications in drug design.