Epik: pKa and Protonation State Prediction through Machine Learning.

Epik version 7 is a software program that uses machine learning for predicting the pKa values and protonation state distribution of complex, druglike molecules. Using an ensemble of atomic graph convolutional neural networks (GCNNs) trained on over 42,000 pKa values across broad chemical space from both experimental and computed origins, the model predicts pKa values with 0.42 and 0.72 pKa unit median absolute and root mean square errors, respectively, across seven test sets. Epik version 7 also generates protonation states and recovers 95% of the most populated protonation states compared to previous versions. Requiring on average only 47 ms per ligand, Epik version 7 is rapid and accurate enough to evaluate protonation states for crucial molecules and prepare ultra-large libraries of compounds to explore vast regions of chemical space. The simplicity and time required for the training allow for the generation of highly accurate models customized to a program's specific chemistry.

Benchmarking Refined and Unrefined AlphaFold2 Structures for Hit Discovery.

The recently developed AlphaFold2 (AF2) algorithm predicts proteins' 3D structures from amino acid sequences. The open AlphaFold protein structure database covers the complete human proteome. Using an industry-leading molecular docking method (Glide), we investigated the virtual screening performance of 37 common drug targets, each with an AF2 structure and known holo and apo structures from the DUD-E data set. In a subset of 27 targets where the AF2 structures are suitable for refinement, the AF2 structures show comparable early enrichment of known active compounds (avg. EF 1%: 13.0) to apo structures (avg. EF 1%: 11.4) while falling behind early enrichment of the holo structures (avg. EF 1%: 24.2). With an induced-fit protocol (IFD-MD), we can refine the AF2 structures using an aligned known binding ligand as the template to improve the performance in structure-based virtual screening (avg. EF 1%: 18.9). Glide-generated docking poses of known binding ligands can also be used as templates for IFD-MD, achieving similar improvements (avg. EF 1% 18.0). Thus, with proper preparation and refinement, AF2 structures show considerable promise for in silico hit identification.

Efficient Exploration of Chemical Space with Docking and Deep Learning.

With the advent of make-on-demand commercial libraries, the number of purchasable compounds available for virtual screening and assay has grown explosively in recent years, with several libraries eclipsing one billion compounds. Today's screening libraries are larger and more diverse, enabling the discovery of more-potent hit compounds and unlocking new areas of chemical space, represented by new core scaffolds. Applying physics-based in silico screening methods in an exhaustive manner, where every molecule in the library must be enumerated and evaluated independently, is increasingly cost-prohibitive. Here, we introduce a protocol for machine learning-enhanced molecular docking based on active learning to dramatically increase throughput over traditional docking. We leverage a novel selection protocol that strikes a balance between two objectives: (1) identifying the best scoring compounds and (2) exploring a large region of chemical space, demonstrating superior performance compared to a purely greedy approach. Together with automated redocking of the top compounds, this method captures almost all the high scoring scaffolds in the library found by exhaustive docking. This protocol is applied to our recent virtual screening campaigns against the D4 and AMPC targets that produced dozens of highly potent, novel inhibitors, and a blind test against the MT1 target. Our protocol recovers more than 80% of the experimentally confirmed hits with a 14-fold reduction in compute cost, and more than 90% of the hit scaffolds in the top 5% of model predictions, preserving the diversity of the experimentally confirmed hit compounds.

AutoQSAR: an automated machine learning tool for best-practice quantitative structure-activity relationship modeling.

AIM: We introduce AutoQSAR, an automated machine-learning application to build, validate and deploy quantitative structure-activity relationship (QSAR) models. METHODOLOGY/RESULTS: The process of descriptor generation, feature selection and the creation of a large number of QSAR models has been automated into a single workflow within AutoQSAR. The models are built using a variety of machine-learning methods, and each model is scored using a novel approach. Effectiveness of the method is demonstrated through comparison with literature QSAR models using identical datasets for six end points: protein-ligand binding affinity, solubility, blood-brain barrier permeability, carcinogenicity, mutagenicity and bioaccumulation in fish. CONCLUSION: AutoQSAR demonstrates similar or better predictive performance as compared with published results for four of the six endpoints while requiring minimal human time and expertise.

WScore: A Flexible and Accurate Treatment of Explicit Water Molecules in Ligand-Receptor Docking.

We have developed a new methodology for protein-ligand docking and scoring, WScore, incorporating a flexible description of explicit water molecules. The locations and thermodynamics of the waters are derived from a WaterMap molecular dynamics simulation. The water structure is employed to provide an atomic level description of ligand and protein desolvation. WScore also contains a detailed model for localized ligand and protein strain energy and integrates an MM-GBSA scoring component with these terms to assess delocalized strain of the complex. Ensemble docking is used to take into account induced fit effects on the receptor conformation, and protein reorganization free energies are assigned via fitting to experimental data. The performance of the method is evaluated for pose prediction, rank ordering of self-docked complexes, and enrichment in virtual screening, using a large data set of PDB complexes and compared with the Glide SP and Glide XP models; significant improvements are obtained.

Docking performance of the glide program as evaluated on the Astex and DUD datasets: a complete set of glide SP results and selected results for a new scoring function integrating WaterMap and glide.

Glide SP mode enrichment results for two preparations of the DUD dataset and native ligand docking RMSDs for two preparations of the Astex dataset are presented. Following a best-practices preparation scheme, an average RMSD of 1.140 Å for native ligand docking with Glide SP is computed. Following the same best-practices preparation scheme for the DUD dataset an average area under the ROC curve (AUC) of 0.80 and average early enrichment via the ROC (0.1 %) metric of 0.12 were observed. 74 and 56 % of the 39 best-practices prepared targets showed AUC over 0.7 and 0.8, respectively. Average AUC was greater than 0.7 for all best-practices protein families demonstrating consistent enrichment performance across a broad range of proteins and ligand chemotypes. In both Astex and DUD datasets, docking performance is significantly improved employing a best-practices preparation scheme over using minimally-prepared structures from the PDB. Enrichment results for WScore, a new scoring function and sampling methodology integrating WaterMap and Glide, are presented for four DUD targets, hivrt, hsp90, cdk2, and fxa. WScore performance in early enrichment is consistently strong and all systems examined show AUC > 0.9 and superior early enrichment to DUD best-practices Glide SP results.

Improving database enrichment through ensemble docking.

While it may seem intuitive that using an ensemble of multiple conformations of a receptor in structure-based virtual screening experiments would necessarily yield improved enrichment of actives relative to using just a single receptor, it turns out that at least in the p38 MAP kinase model system studied here, a very large majority of all possible ensembles do not yield improved enrichment of actives. However, there are combinations of receptor structures that do lead to improved enrichment results. We present here a method to select the ensembles that produce the best enrichments that does not rely on knowledge of active compounds or sophisticated analyses of the 3D receptor structures. In the system studied here, the small fraction of ensembles of up to 3 receptors that do yield good enrichments of actives were identified by selecting ensembles that have the best mean GlideScore for the top 1% of the docked ligands in a database screen of actives and drug-like "decoy" ligands. Ensembles of two receptors identified using this mean GlideScore metric generally outperform single receptors, while ensembles of three receptors identified using this metric consistently give optimal enrichment factors in which, for example, 40% of the known actives outrank all the other ligands in the database.

Flexible ligand docking with Glide.

Glide is a ligand docking program for predicting protein-ligand binding modes and ranking ligands via high-throughput virtual screening. Glide utilizes two different scoring functions, SP and XP GlideScore, to rank-order compounds. Three modes of sampling ligand conformational and positional degrees of freedom are available to determine the optimal ligand orientation relative to a rigid protein receptor geometry. This unit presents protocols for flexible ligand docking with Glide, optionally including ligand constraints or ligand molecular similarities.

Extra precision glide: docking and scoring incorporating a model of hydrophobic enclosure for protein-ligand complexes.

A novel scoring function to estimate protein-ligand binding affinities has been developed and implemented as the Glide 4.0 XP scoring function and docking protocol. In addition to unique water desolvation energy terms, protein-ligand structural motifs leading to enhanced binding affinity are included: (1) hydrophobic enclosure where groups of lipophilic ligand atoms are enclosed on opposite faces by lipophilic protein atoms, (2) neutral-neutral single or correlated hydrogen bonds in a hydrophobically enclosed environment, and (3) five categories of charged-charged hydrogen bonds. The XP scoring function and docking protocol have been developed to reproduce experimental binding affinities for a set of 198 complexes (RMSDs of 2.26 and 1.73 kcal/mol over all and well-docked ligands, respectively) and to yield quality enrichments for a set of fifteen screens of pharmaceutical importance. Enrichment results demonstrate the importance of the novel XP molecular recognition and water scoring in separating active and inactive ligands and avoiding false positives.

Integrated Modeling Program, Applied Chemical Theory (IMPACT).

We provide an overview of the IMPACT molecular mechanics program with an emphasis on recent developments and a description of its current functionality. With respect to core molecular mechanics technologies we include a status report for the fixed charge and polarizable force fields that can be used with the program and illustrate how the force fields, when used together with new atom typing and parameter assignment modules, have greatly expanded the coverage of organic compounds and medicinally relevant ligands. As we discuss in this review, explicit solvent simulations have been used to guide our design of implicit solvent models based on the generalized Born framework and a novel nonpolar estimator that have recently been incorporated into the program. With IMPACT it is possible to use several different advanced conformational sampling algorithms based on combining features of molecular dynamics and Monte Carlo simulations. The program includes two specialized molecular mechanics modules: Glide, a high-throughput docking program, and QSite, a mixed quantum mechanics/molecular mechanics module. These modules employ the IMPACT infrastructure as a starting point for the construction of the protein model and assignment of molecular mechanics parameters, but have then been developed to meet specialized objectives with respect to sampling and the energy function.

Glide: a new approach for rapid, accurate docking and scoring. 1. Method and assessment of docking accuracy.

Unlike other methods for docking ligands to the rigid 3D structure of a known protein receptor, Glide approximates a complete systematic search of the conformational, orientational, and positional space of the docked ligand. In this search, an initial rough positioning and scoring phase that dramatically narrows the search space is followed by torsionally flexible energy optimization on an OPLS-AA nonbonded potential grid for a few hundred surviving candidate poses. The very best candidates are further refined via a Monte Carlo sampling of pose conformation; in some cases, this is crucial to obtaining an accurate docked pose. Selection of the best docked pose uses a model energy function that combines empirical and force-field-based terms. Docking accuracy is assessed by redocking ligands from 282 cocrystallized PDB complexes starting from conformationally optimized ligand geometries that bear no memory of the correctly docked pose. Errors in geometry for the top-ranked pose are less than 1 A in nearly half of the cases and are greater than 2 A in only about one-third of them. Comparisons to published data on rms deviations show that Glide is nearly twice as accurate as GOLD and more than twice as accurate as FlexX for ligands having up to 20 rotatable bonds. Glide is also found to be more accurate than the recently described Surflex method.

Testing electronic structure methods for describing intermolecular H...H interactions in supramolecular chemistry.

In this article a wide variety of computational approaches (molecular mechanics force fields, semiempirical formalisms, and hybrid methods, namely ONIOM calculations) have been used to calculate the energy and geometry of the supramolecular system 2-(2'-hydroxyphenyl)-4-methyloxazole (HPMO) encapsulated in beta-cyclodextrin (beta-CD). The main objective of the present study has been to examine the performance of these computational methods when describing the short range H. H intermolecular interactions between guest (HPMO) and host (beta-CD) molecules. The analyzed molecular mechanics methods do not provide unphysical short H...H contacts, but it is obvious that their applicability to the study of supramolecular systems is rather limited. For the semiempirical methods, MNDO is found to generate more reliable geometries than AM1, PM3 and the two recently developed schemes PDDG/MNDO and PDDG/PM3. MNDO results only give one slightly short H...H distance, whereas the NDDO formalisms with modifications of the Core Repulsion Function (CRF) via Gaussians exhibit a large number of short to very short and unphysical H...H intermolecular distances. In contrast, the PM5 method, which is the successor to PM3, gives very promising results. Our ONIOM calculations indicate that the unphysical optimized geometries from PM3 are retained when this semiempirical method is used as the low level layer in a QM:QM formulation. On the other hand, ab initio methods involving good enough basis sets, at least for the high level layer in a hybrid ONIOM calculation, behave well, but they may be too expensive in practice for most supramolecular chemistry applications. Finally, the performance of the evaluated computational methods has also been tested by evaluating the energetic difference between the two most stable conformations of the host(beta-CD)-guest(HPMO) system.

Extension of the PDDG/PM3 and PDDG/MNDO semiempirical molecular orbital methods to the halogens.

The new semiempirical methods, PDDG/PM3 and PDDG/MNDO, have been parameterized for halogens. For comparison, the original MNDO and PM3 were also reoptimized for the halogens using the same training set; these modified methods are referred to as MNDO' and PM3'. For 442 halogen-containing molecules, the smallest mean absolute error (MAE) in heats of formation is obtained with PDDG/PM3 (5.6 kcal/mol), followed by PM3' (6.1 kcal/mol), PDDG/MNDO (6.6 kcal/mol), PM3 (8.1 kcal/mol), MNDO' (8.5 kcal/mol), AM1 (11.1 kcal/mol), and MNDO (14.0 kcal/mol). For normal-valent halogen-containing molecules, the PDDG methods also provide improved heats of formation over MNDO/d. Hypervalent compounds were not included in the training set and improvements over the standard NDDO methods with sp basis sets were not obtained. For small haloalkanes, the PDDG methods yield more accurate heats of formation than are obtained from density functional theory (DFT) with the B3LYP and B3PW91 functionals using large basis sets. PDDG/PM3 and PM3' also give improved binding energies over the standard NDDO methods for complexes involving halide anions, and they are competitive with B3LYP/6-311++G(d,p) results including thermal corrections. Among the semiempirical methods studied, PDDG/PM3 also generates the best agreement with high-level ab initio G2 and CCSD(T) intrinsic activation energies for S(N)2 reactions involving methyl halides and halide anions. Finally, the MAEs in ionization potentials, dipole moments, and molecular geometries show that the parameter sets for the PDDG and reoptimized NDDO methods reduce the MAEs in heats of formation without compromising the other important QM observables.

Contributions of conformational compression and preferential transition state stabilization to the rate enhancement by chorismate mutase.

The rate enhancement provided by the chorismate mutase (CM) enzyme for the Claisen rearrangement of chorismate to prephenate has been investigated by application of the concept of near attack conformations (NACs). Using a combined QM/MM Monte Carlo/free-energy perturbation (MC/FEP) method, 82% and 100% of chorismate conformers were found to be NAC structures in water and in the CM active site, respectively. Consequently, the conversion of non-NACs to NACs does not contribute to the free energy of activation from preorganization of the substrate into NACs. The FEP calculations yielded differences in free energies of activation that well reproduce the experimental data. Additional calculations indicate that the rate enhancement by CM over the aqueous phase results primarily from conformational compression of NACs by the enzyme and that this process is enthalpically controlled. This suggests that preferential stabilization of the transition state in the enzyme environment relative to water plays a secondary role in the catalysis by CM.

Investigation of solvent effects for the Claisen rearrangement of chorismate to prephenate: mechanistic interpretation via near attack conformations.

Solvent effects on the rate of the Claisen rearrangement of chorismate to prephenate have been examined in water and methanol. The preequilibrium free-energy differences between diaxial and diequatorial conformers of chorismate, which had previously been implicated as the sole basis for the observed 100-fold rate increase in water over methanol, have been reframed using the near attack conformation (NAC) concept of Bruice and co-workers. Using a combined QM/MM Monte Carlo/free-energy perturbation (MC/FEP) method, 82%, 57%, and 1% of chorismate conformers were found to be NAC structures (NACs) in water, methanol, and the gas phase, respectively. As a consequence, the conversion of non-NACs to NACs provides no free-energy contributions to the overall relative reaction rates in water versus methanol. Free-energy perturbation calculations yielded differences in free energies of activation for the two polar protic solvents and the gas phase. The rate enhancement in water over the gas phase arises from preferential hydration of the transition state (TS) relative to the reactants via increased hydrogen bonding and long-range electrostatic interactions, which accompany bringing the two negatively charged carboxylates into closer proximity. More specifically, there is an increase of 1.3 and 0.6 hydrogen bonds to the carboxylate groups and the ether oxygen, respectively, in going from the reactant to the TS in water. In methanol, the corresponding changes in hydrogen bonding with first shell solvent molecules are small; the rate enhancement arises primarily from the enhanced long-range interactions with solvent molecules. Thus, the reaction occurs faster in water than in methanol due to greater stabilization of the TS in water by specific interactions with first shell solvent molecules.

PDDG/PM3 and PDDG/MNDO: improved semiempirical methods.

Two new semiempirical methods employing a Pairwise Distance Directed Gaussian modification have been developed: PDDG/PM3 and PDDG/MNDO; they are easily implemented in existing software, and yield heats of formation for compounds containing C, H, N, and O atoms with significantly improved accuracy over the standard NDDO schemes, PM5, PM3, AM1, and MNDO. The PDDG/PM3 results for heats of formation also show substantial improvement over density functional theory with large basis sets. The PDDG modifications consist of a single function, which is added to the existing pairwise core repulsion functions within PM3 and MNDO, a reparameterized semiempirical parameter set, and modified computation of the energy of formation of a gaseous atom. The PDDG addition introduces functional group information via pairwise atomic interactions using only atom-based parameters. For 622 diverse molecules containing C, H, N, and O atoms, mean absolute errors in calculated heats of formation are reduced from 4.4 to 3.2 kcal/mol and from 8.4 to 5.2 kcal/mol using the PDDG modified versions of PM3 and MNDO over the standard versions, respectively. Several specific problems are overcome, including the relative stability of hydrocarbon isomers, and energetics of small rings and molecules containing multiple heteroatoms. The internal consistency of PDDG energies is also significantly improved, enabling more reliable analysis of isomerization energies and trends across series of molecules; PDDG isomerization energies show significant improvement over B3LYP/6-31G* results. Comparison of heats of formation, ionization potentials, dipole moments, isomer, and conformer energetics, intermolecular interaction energies, activation energies, and molecular geometries from the PDDG techniques is made to experimental data and values from other semiempirical and ab initio methods.

Improved semiempirical heats of formation through the use of bond and group equivalents.

Deficiencies in energetics obtained using the common semiempirical methods, AM1, PM3, and MNDO, may partly be traced to the use of pseudoatomic equivalents for conversion of molecular energies to heats of formation at 298 K. We present an alternative scheme based on the use of bond and group equivalents. Values for the 61 bond and group equivalents necessary for treatment of molecules containing the common organic elements, hydrogen, carbon, nitrogen, and oxygen have been derived. For a set of 583 neutral, closed-shell molecules mean absolute errors in AM1, PM3, and MNDO heats of formation are reduced from 6.6, 4.2, and 8.2 kcal/mol to 2.3, 2.2, and 3.0 kcal/mol, respectively. Several systematic problems are overcome in the present scheme including relative stabilities of branched hydrocarbons, energetics of conjugated systems, heats of formation of long chain hydrocarbons, and enthalpies of molecules containing multiple heteroatoms. Although the approach is restricted to molecules with well-defined functional groups, the equivalents are easy to incorporate and are chemically relevant. This revised procedure allows semiempirical methods to be used for far more reliable evaluations of heats of reactions. Estimates are made of the errors inherent in these semiempirical formalisms, arising from integral approximations and the neglect of explicit treatment of electron correlation effects, while excluding those from inadequate parameterization.