Enhancing Hit Discovery in Virtual Screening through Absolute Protein-Ligand Binding Free-Energy Calculations.

In the hit identification stage of drug discovery, a diverse chemical space needs to be explored to identify initial hits. Contrary to empirical scoring functions, absolute protein-ligand binding free-energy perturbation (ABFEP) provides a theoretically more rigorous and accurate description of protein-ligand binding thermodynamics and could, in principle, greatly improve the hit rates in virtual screening. In this work, we describe an implementation of an accurate and reliable ABFEP method in FEP+. We validated the ABFEP method on eight congeneric compound series binding to eight protein receptors including both neutral and charged ligands. For ligands with net charges, the alchemical ion approach is adopted to avoid artifacts in electrostatic potential energy calculations. The calculated binding free energies correlate with experimental results with a weighted average of R2 = 0.55 for the entire dataset. We also observe an overall root-mean-square error (RMSE) of 1.1 kcal/mol after shifting the zero-point of the simulation data to match the average experimental values. Through ABFEP calculations using apo versus holo protein structures, we demonstrated that the protein conformational and protonation state changes between the apo and holo proteins are the main physical factors contributing to the protein reorganization free energy manifested by the overestimation of raw ABFEP calculated binding free energies using the holo structures of the proteins. Furthermore, we performed ABFEP calculations in three virtual screening applications for hit enrichment. ABFEP greatly improves the hit rates as compared to docking scores or other methods like metadynamics. The good performance of ABFEP in rank ordering compounds demonstrated in this work confirms it as a useful tool to improve the hit rates in virtual screening, thus facilitating hit discovery.

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.

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.

Shape-Based Virtual Screening of a Billion-Compound Library Identifies Mycobacterial Lipoamide Dehydrogenase Inhibitors.

Lpd (lipoamide dehydrogenase) in Mycobacterium tuberculosis (Mtb) is required for virulence and is a genetically validated tuberculosis (TB) target. Numerous screens have been performed over the last decade, yet only two inhibitor series have been identified. Recent advances in large-scale virtual screening methods combined with make-on-demand compound libraries have shown the potential for finding novel hits. In this study, the Enamine REAL library consisting of ∼1.12 billion compounds was efficiently screened using the GPU Shape screen method against Mtb Lpd to find additional chemical matter that would expand on the known sulfonamide inhibitor series. We identified six new inhibitors with IC50 in the range of 5-100 µM. While these compounds remained chemically close to the already known sulfonamide series inhibitors, some diversity was found in the cores of the hits. The two most potent hits were further validated by one-step potency optimization to submicromolar levels. The co-crystal structure of optimized analogue TDI-13537 provided new insights into the potency determinants of the series.

Strained cyclophane macrocycles: impact of progressive ring size reduction on synthesis and structure.

The synthesis, X-ray crystal structures, and calculated strain energies are reported for a homologous series of 11- to 14-membered drug-like cyclophane macrocycles, representing an unusual region of chemical space that can be difficult to access synthetically. The ratio of macrocycle to dimer, generated via a copper catalyzed azide-alkyne cycloaddition macrocyclization in flow at elevated temperature, could be rationalized in terms of the strain energy in the macrocyclic product. The progressive increase in strain resulting from reduction in macrocycle ring size, or the introduction of additional conformational constraints, results in marked deviations from typical geometries. These strained cyclophane macrocyclic systems provide access to spatial orientations of functionality that would not be readily available in unstrained or acyclic analogs. The most strained system prepared represents the first report of an 11-membered cyclophane containing a 1,4-disubstituted 1,2,3-triazole ring and establishes a limit to the ring strain that can be generated using this macrocycle synthesis methodology.

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.

Successful application of the DBLOC method to the hydroxylation of camphor by cytochrome p450.

The activation barrier for the hydroxylation of camphor by cytochrome P450 was computed using a mixed quantum mechanics/molecular mechanics (QM/MM) model of the full protein-ligand system and a fully QM calculation using a cluster model of the active site at the B3LYP/LACVP*/LACV3P** level of theory, which consisted of B3LYP/LACV3P** single point energies computed at B3LYP/LACVP* optimized geometries. From the QM/MM calculation, a barrier height of 17.5 kcal/mol was obtained, while the experimental value was known to be less than or equal to 10 kcal/mol. This process was repeated using the D3 correction for hybrid DFT in order to investigate whether the inadequate treatment of dispersion interaction was responsible for the overestimation of the barrier. While the D3 correction does reduce the computed barrier to 13.3 kcal/mol, it was still in disagreement with experiment. After application of a series of transition metal optimized localized orbital corrections (DBLOC) and without any refitting of parameters, the barrier was further reduced to 10.0 kcal/mol, which was consistent with the experimental results. The DBLOC method to CH bond activation in methane monooxygenase (MMO) was also applied, as a second, independent test. The barrier in MMO was known, by experiment, to be 15.4 kcal/mol. After application of the DBLOC corrections to the MMO barrier compute by B3LYP, in a previous study, and accounting for dispersion with Grimme's D3 method, the unsigned deviation from experiment was improved from 3.2 to 2.3 kcal/mol. These results suggested that the combination of dispersion plus localized orbital corrections could yield significant quantitative improvements in modeling the catalytic chemistry of transition-metal containing enzymes, within the limitations of the statistical errors of the model, which appear to be on the order of approximately 2 kcal/mole.

Evaluation of the Performance of the B3LYP, PBE0, and M06 DFT Functionals, and DBLOC-Corrected Versions, in the Calculation of Redox Potentials and Spin Splittings for Transition Metal Containing Systems.

We have evaluated the performance of the M06 and PBE0 functionals in their ability to calculate spin splittings and redox potentials for octahedral complexes containing a first transition metal series atom. The mean unsigned errors (MUEs) for these two functionals are similar to those obtained for B3LYP using the same data sets. We then apply our localized orbital correction approach for transition metals, DBLOC, in an effort to improve the results obtained with both functionals. The PBE0-DBLOC results are remarkably close in both MUE and parameter values to those obtained for the B3LYP-DBLOC method. The M06-DBLOC results are less accurate, but the parameter values and trends are still qualitatively very similar. These results demonstrate that DBLOC corrected methods are substantially more accurate for these systems than any of the uncorrected functionals we have tested and that the deviations between hybrid DFT methods and experiment for transition metal containing systems exhibit striking physically based regularities which are very similar for the three functionals that we have examined, despite significant differences in the details of each model.

Accurate pKa prediction in first-row hexaaqua transition metal complexes using the B3LYP-DBLOC method.

Acid dissociation constants are computed with density functional theory (DFT) for a series of ten first-row octahedral hexaaqua transition metal complexes at the B3LYP/LACV3P** level of theory. These results are then scaled, primarily to correct for basis set effects (as in previous work on predicting pKa's in organic systems1-5). Finally, localized orbital corrections (LOCs), developed by fitting properties such as ionization potentials, electron affinities, and ligand removal energies in prior publications,3,4,6,7 are applied without any further parameter adjustment. The combination of a single scale factor with the DBLOC (localized orbital corrections for first row transition metals) corrections (and thus a single adjustable parameter in all) improves the mean unsigned error from 5.7 pKa units (with no parameters) to 0.9 pKa units (maximum error 2.2 pKa units), which is close to chemical accuracy for this type of system. These results provide further encouragement with regard to the ability of the B3LYP-DBLOC model to provide accurate and robust results for DFT calculations on transition metal containing species.

Charge Model 5: An Extension of Hirshfeld Population Analysis for the Accurate Description of Molecular Interactions in Gaseous and Condensed Phases.

We propose a novel approach to deriving partial atomic charges from population analysis. The new model, called Charge Model 5 (CM5), yields class IV partial atomic charges by mapping from those obtained by Hirshfeld population analysis of density functional electronic charge distributions. The CM5 model utilizes a single set of parameters derived by fitting to reference values of the gas-phase dipole moments of 614 molecular structures. An additional test set (not included in the CM5 parametrization) contained 107 singly charged ions with nonzero dipole moments, calculated from the accurate electronic charge density, with respect to the center of nuclear charges. The CM5 model is applicable to any charged or uncharged molecule composed of any element of the periodic table in the gas phase or in solution. The CM5 model predicts dipole moments for the tested molecules that are more accurate on average than those from the original Hirshfeld method or from many other popular schemes including atomic polar tensor and Löwdin, Mulliken, and natural population analyses. In addition, the CM5 charge model is essentially independent of a basis set. It can be used with larger basis sets, and thereby this model significantly improves on our previous charge models CMx (x = 1-4 or 4M) and other methods that are prone to basis set sensitivity. CM5 partial atomic charges are less conformationally dependent than those derived from electrostatic potentials. The CM5 model does not suffer from ill conditioning for buried atoms in larger molecules, as electrostatic fitting schemes sometimes do. The CM5 model can be used with any level of electronic structure theory (Hartree-Fock, post-Hartree-Fock, and other wave function correlated methods or density functional theory) as long as an accurate electronic charge distribution and a Hirshfeld analysis can be computed for that level of theory.

Why a proximity-induced Diels-Alder reaction is so fast.

Unlike normal Diels-Alder reactions of acyclic alkadienes with alkenes, the vinylbicyclo[2.2.2]octene employed in the Baran total synthesis of vinigrol undergoes a quantitative Diels-Alder reaction with a tethered alkene at room temperature. Density functional theory calculations reveal that this unprecedented reactivity originates from a combination of preorganization, diene strain, and tether stabilization.