Conformational energy range of ligands in protein crystal structures: The difficult quest for accurate understanding

Applying atomistic neural networks to bias conformer ensembles towards bioactive-like conformations

Identifying bioactive conformations of small molecules is an essential process for virtual screening applications relying on three-dimensional structure such as molecular docking. For most small molecules, conformer generators retrieve at least one bioactive-like conformation, with an atomic root-mean-square deviation (ARMSD) lower than 1 Å, among the set of low-energy conformers generated. However, there is currently no general method to prioritise these likely target-bound conformations in the ensemble. In this work, we trained atomistic neural networks (AtNNs) on 3D information of generated conformers of a curated subset of PDBbind ligands to predict the ARMSD to their closest bioactive conformation, and evaluated the early enrichment of bioactive-like conformations when ranking conformers by AtNN prediction. AtNN ranking was compared with bioactivity-unaware baselines such as ascending Sage force field energy ranking, and a slower bioactivity-based baseline ranking by ascending Torsion Fingerprint Deviation to the Maximum Common Substructure to the most similar molecule in the training set (TFD2SimRefMCS). On test sets from random ligand splits of PDBbind, ranking conformers using ComENet, the AtNN encoding the most 3D information, leads to early enrichment of bioactive-like conformations with a median BEDROC of 0.29 ± 0.02, outperforming the best bioactivity-unaware Sage energy ranking baseline (median BEDROC of 0.18 ± 0.02), and performing on a par with the bioactivity-based TFD2SimRefMCS baseline (median BEDROC of 0.31 ± 0.02). The improved performance of the AtNN and TFD2SimRefMCS baseline is mostly observed on test set ligands that bind proteins similar to proteins observed in the training set. On a more challenging subset of flexible molecules, the bioactivity-unaware baselines showed median BEDROCs up to 0.02, while AtNNs and TFD2SimRefMCS showed median BEDROCs between 0.09 and 0.13. When performing rigid ligand re-docking of PDBbind ligands with GOLD using the 1% top-ranked conformers, ComENet ranked conformers showed a higher successful docking rate than bioactivity-unaware baselines, with a rate of 0.48 ± 0.02 compared to CSD probability baseline with a rate of 0.39 ± 0.02. Similarly, on a pharmacophore searching experiment, selecting the 20% top-ranked conformers ranked by ComENet showed higher hit rate compared to baselines. Hence, the approach presented here uses AtNNs successfully to focus conformer ensembles towards bioactive-like conformations, representing an opportunity to reduce computational expense in virtual screening applications on known targets that require input conformations.

Study of SQ109 analogs binding to mycobacterium MmpL3 transporter using MD simulations and alchemical relative binding free energy calculations

N-geranyl-N΄-(2-adamantyl)ethane-1,2-diamine (SQ109) is a tuberculosis drug that has high potency against Mycobacterium tuberculosis (Mtb) and may function by blocking cell wall biosynthesis. After the crystal structure of MmpL3 from Mycobacterium smegmatis in complex with SQ109 became available, it was suggested that SQ109 inhibits Mmpl3 mycolic acid transporter. Here, we showed using molecular dynamics (MD) simulations that the binding profile of nine SQ109 analogs with inhibitory potency against Mtb and alkyl or aryl adducts at C-2 or C-1 adamantyl carbon to MmpL3 was consistent with the X-ray structure of MmpL3 - SQ109 complex. We showed that rotation of SQ109 around carbon-carbon bond in the monoprotonated ethylenediamine unit favors two gauche conformations as minima in water and lipophilic solvent using DFT calculations as well as inside the transporter's binding area using MD simulations. The binding assays in micelles suggested that the binding affinity of the SQ109 analogs was increased for the larger, more hydrophobic adducts, which was consistent with our results from MD simulations of the SQ109 analogues suggesting that sizeable C-2 adamantyl adducts of SQ109 can fill a lipophilic region between Y257, Y646, F260 and F649 in MmpL3. This was confirmed quantitatively by our calculations of the relative binding free energies using the thermodynamic integration coupled with MD simulations method with a mean assigned error of 0.74 kcal mol^-1 compared to the experimental values.

Rhenium(V) Complexes as Cysteine-Targeting Coordinate Covalent Warheads

Interest in covalent enzyme inhibitors as therapeutic agents has seen a recent resurgence. Covalent enzyme inhibitors typically possess an organic functional group that reacts with a key feature of the target enzyme, often a nucleophilic cysteine residue. Herein, the application of small, modular Re^V complexes as inorganic cysteine-targeting warheads is described. These metal complexes were found to react with cysteine residues rapidly and selectively. To demonstrate the utility of these Re^V complexes, their reactivity with SARS-CoV-2-associated cysteine proteases is presented, including the SARS-CoV-2 main protease and papain-like protease and human enzymes cathepsin B and L. As all of these proteins are cysteine proteases, these enzymes were found to be inhibited by the Re^V complexes through the formation of adducts. These findings suggest that these Re^V complexes could be used as a new class of warheads for targeting surface accessible cysteine residues in disease-relevant target proteins.

Collaborative Assessment of Molecular Geometries and Energies from the Open Force Field

Force fields form the basis for classical molecular simulations, and their accuracy is crucial for the quality of, for instance, protein-ligand binding simulations in drug discovery. The huge diversity of small-molecule chemistry makes it a challenge to build and parameterize a suitable force field. The Open Force Field Initiative is a combined industry and academic consortium developing a state-of-the-art small-molecule force field. In this report, industry members of the consortium worked together to objectively evaluate the performance of the force fields (referred to here as OpenFF) produced by the initiative on a combined public and proprietary dataset of 19,653 relevant molecules selected from their internal research and compound collections. This evaluation was important because it was completely blind; at most partners, none of the molecules or data were used in force field development or testing prior to this work. We compare the Open Force Field "Sage" version 2.0.0 and "Parsley" version 1.3.0 with GAFF-2.11-AM1BCC, OPLS4, and SMIRNOFF99Frosst. We analyzed force-field-optimized geometries and conformer energies compared to reference quantum mechanical data. We show that OPLS4 performs best, and the latest Open Force Field release shows a clear improvement compared to its predecessors. The performance of established force fields such as GAFF-2.11 was generally worse. While OpenFF researchers were involved in building the benchmarking infrastructure used in this work, benchmarking was done entirely in-house within industrial organizations and the resulting assessment is reported here. This work assesses the force field performance using separate benchmarking steps, external datasets, and involving external research groups. This effort may also be unique in terms of the number of different industrial partners involved, with 10 different companies participating in the benchmark efforts.

An Efficient Approach to Large-Scale Ab Initio Conformational Energy Profiles of Small Molecules

Accurate conformational energetics of molecules are of great significance to understand maby chemical properties. They are also fundamental for high-quality parameterization of force fields. Traditionally, accurate conformational profiles are obtained with density functional theory (DFT) methods. However, obtaining a reliable energy profile can be time-consuming when the molecular sizes are relatively large or when there are many molecules of interest. Furthermore, incorporation of data-driven deep learning methods into force field development has great requirements for high-quality geometry and energy data. To this end, we compared several possible alternatives to the traditional DFT methods for conformational scans, including the semi-empirical method GFN2-xTB and the neural network potential ANI-2x. It was found that a sequential protocol of geometry optimization with the semi-empirical method and single-point energy calculation with high-level DFT methods can provide satisfactory conformational energy profiles hundreds of times faster in terms of optimization.

Computational design of a sensitive, selective phase-changing sensor protein for the VX nerve agent

The VX nerve agent is one of the deadliest chemical warfare agents. Specific, sensitive, real-time detection methods for this neurotoxin have not been reported. The creation of proteins that use biological recognition to fulfill these requirements using directed evolution or library screening methods has been hampered because its toxicity makes laboratory experimentation extraordinarily expensive. A pair of VX-binding proteins were designed using a supercharged scaffold that couples a large-scale phase change from unstructured to folded upon ligand binding, enabling fully internal binding sites that present the maximum surface area possible for high affinity and specificity in target recognition. Binding site residues were chosen using a new distributed evolutionary algorithm implementation in protCAD. Both designs detect VX at parts per billion concentrations with high specificity. Computational design of fully buried molecular recognition sites, in combination with supercharged phase-changing chassis proteins, enables the ready development of a new generation of small-molecule biosensors.

Increased Molecular Flexibility Widens the Gap between K i and K d values in Screening for Retinoid X Receptor Modulators

Screening for small-molecule modulators targeting a particular receptor is frequently based on measurement of K _d, i.e., the binding constant between the receptor and the compound of interest. However, K _d values also reflect binding at receptor protein sites other than the modulatory site. We designed derivatives of retinoid X receptor (RXR) antagonist CBTF-EE (1) with modifications that altered their conformational flexibility. Compounds 6a,b and 7a,b showed quite similar K _d values, but 7a,b exhibited 10-fold higher K _i values than those of 6a,b. Further, 6a,b showed potent RXR-antagonistic activity, while 7a,b were inactive. These results suggest that increased conformational flexibility promotes binding at nontarget receptor sites. In this situation, conventional determination of K _d is less effective for screening purposes than the determination of K _i using a ligand that binds specifically to the site regulating transcriptional activity. Thus, the use of K _i values for orthosteric ligands may increase the hit rate in screening active regulatory molecules.

Ligand Strain Energy in Large Library Docking

While small molecule internal strain is crucial to molecular docking, using it in evaluating ligand scores has remained elusive. Here, we investigate a technique that calculates strain using relative torsional populations in the Cambridge Structural Database, enabling fast precalculation of these energies. In retrospective studies of large docking screens of the dopamine D4 receptor and of AmpC β-lactamase, where close to 600 docking hits were tested experimentally, including such strain energies improved hit rates by preferentially reducing the ranks of strained high-scoring decoy molecules. In a 40-target subset of the DUD-E benchmark, we found two thresholds that usefully distinguished between ligands and decoys: one based on the total strain energy of the small molecules and another based on the maximum strain allowed for any given torsion within them. Using these criteria, about 75% of the benchmark targets had improved enrichment after strain filtering. Relying on precalculated population distributions, this approach is rapid, taking less than 0.04 s to evaluate a conformation on a standard core, making it pragmatic for precalculating strain in even ultralarge libraries. Since it is scoring function agnostic, it may be useful to multiple docking approaches; it is openly available at http://tldr.docking.org.

The Good, the Bad, and the Twisted Revisited: An Analysis of Ligand Geometry in Highly Resolved Protein-Ligand X-ray Structures

An analysis of the rotatable bond geometry of drug-like ligand models is reported for high-resolution (<1.1 Å) crystallographic protein-ligand complexes. In cases where the ligand fit to the electron density is very good, unusual torsional geometry is rare and, most often, though not exclusively, associated with strong polar, metal, or covalent ligand-protein interactions. It is rarely associated with a torsional strain of greater than 2 kcal mol^-1 by calculation. An unusual torsional geometry is more prevalent where the fit to electron density is not perfect. Multiple low-strain conformer bindings were observed in 21% of the set and, it is suggested, may also lie behind many of the 35% of single-occupancy cases, where a poor fit to the e-density was found. It is concluded that multiple conformer ligand binding is an under-recognized phenomenon in structure-based drug design and that there is a need for more robust crystallographic refinement methods to better handle such cases.

In silico drug discovery for a complex immunotherapeutic target - human c-Rel protein

Target evaluation and rational drug design rely on identifying and characterising small-molecule binding sites on therapeutically relevant target proteins. Immunotherapeutics development is especially challenging because of complex disease etiology and heterogenous nature of targets. c-Rel protein, a promising target in many human inflammatory and cancer pathologies, was selected as a case study for an effective in silico screening platform development since this transcription factor currently has no successful therapeutic inhibitors or modulators. This study introduces a novel in silico screening approach to probe binding sites using structural validation sets, molecular modelling and describes a method of a computer-aided drug design when a crystal structure is not available for the target of interest. In addition, we showed that binding sites can be analysed with the machine learning as well as molecular simulation approaches to help assess and systematically analyse how drug candidates can exert their mode of action. Finally, this cutting-edge approach was subjected to a high through-put virtual screen of selected 34 M drug-like compounds filtered from a library of 659 M compounds by identifying the most promising structures and proposing potential action mechanisms for the future development of highly selective human c-Rel inhibitors and/or modulators.

Macromolecular refinement of X-ray and cryoelectron microscopy structures with Phenix/OPLS3e for improved structure and ligand quality

With the advent of the resolution revolution in cryoelectron microscopy (cryo-EM), low-resolution refinement is common, and likewise increases the need for a reliable force field. Here, we report on the incorporation of the OPLS3e force field with the VSGB2.1 solvation model in the structure determination package Phenix. Our results show significantly improved structure quality and reduced ligand strain at lower resolution for X-ray refinement. For refinement of cryo-EM-based structures, we find comparable quality structures, goodness-of-fit, and reduced ligand strain. We also show how structure quality and ligand strain are related to the map-model cross-correlation as a function of data weight, and how that can detect overfitting. Signs of overfitting are found in over half of our cryo-EM dataset, which can be remedied by a re-refinement at a lower data weight. Finally, a start-to-end script for refining structures with Phenix/OPLS3e is available in the Schrödinger 2020-3 distribution.

Conformational Strain of Macrocyclic Peptides in Ligand-Receptor Complexes Based on Advanced Refinement of Bound-State Conformers

Macrocyclic peptides are an important modality in drug discovery, but molecular design is limited due to the complexity of their conformational landscape. To better understand conformational propensities, global strain energies were estimated for 156 protein-macrocyclic peptide cocrystal structures. Unexpectedly large strain energies were observed when the bound-state conformations were modeled with positional restraints. Instead, low-energy conformer ensembles were generated using xGen that fit experimental X-ray electron density maps and gave reasonable strain energy estimates. The ensembles featured significant conformational adjustments while still fitting the electron density as well or better than the original coordinates. Strain estimates suggest the interaction energy in protein-ligand complexes can offset a greater amount of strain for macrocyclic peptides than for small molecules and non-peptidic macrocycles. Across all molecular classes, the approximate upper bound on global strain energies had the same relationship with molecular size, and bound-state ensembles from xGen yielded favorable binding energy estimates.

Conformational analysis of macrocycles: comparing general and specialized methods

Abstract

Macrocycles represent an important class of medicinally relevant small molecules due to their interesting biological properties. Therefore, a firm understanding of their conformational preferences is important for drug design. Given the importance of macrocycle-protein modelling in drug discovery, we envisaged that a systematic study of both classical and recent specialized methods would provide guidance for other practitioners within the field. In this study we compare the performance of the general, well established conformational analysis methods Monte Carlo Multiple Minimum (MCMM) and Mixed Torsional/Low-Mode sampling (MTLMOD) with two more recent and specialized macrocycle sampling techniques: MacroModel macrocycle Baseline Search (MD/LLMOD) and Prime macrocycle conformational sampling (PRIME-MCS). Using macrocycles extracted from 44 macrocycle-protein X-ray crystallography complexes, we evaluated each method based on their ability to (i) generate unique conformers, (ii) generate unique macrocycle ring conformations, (iii) identify the global energy minimum, (iv) identify conformers similar to the X-ray ligand conformation after Protein Preparation Wizard treatment (X-ray_ppw), and (v) to the X-ray_ppw ring conformation. Computational speed was also considered. In addition, conformational coverage, as defined by the number of conformations identified, was studied. In order to study the relative energies of the bioactive conformations, the energy differences between the global energy minima and the energy minimized X-ray_ppw structures and, the global energy minima and the MCMM-Exhaustive (1,000,000 search steps) generated conformers closest to the X-ray_ppw structure, were calculated and analysed. All searches were performed using relatively short run times (10,000 steps for MCMM, MTLMOD and MD/LLMOD). To assess the performance of the methods, they were compared to an exhaustive MCMM search using 1,000,000 search steps for each of the 44 macrocycles (requiring ca 200 times more CPU time). Prior to our analysis, we also investigated if the general search methods MCMM and MTLMOD could also be optimized for macrocycle conformational sampling. Taken together, our work concludes that the more general methods can be optimized for macrocycle modelling by slightly adjusting the settings around the ring closure bond. In most cases, MCMM and MTLMOD with either standard or enhanced settings performed well in comparison to the more specialized macrocycle sampling methods MD/LLMOD and PRIME-MCS. When using enhanced settings for MCMM and MTLMOD, the X-ray_ppw conformation was regenerated with the greatest accuracy. The, MD/LLMOD emerged as the most efficient method for generating the global energy minima.

Graphic abstract

Electronic supplementary material

The online version of this article (10.1007/s10822-020-00277-2) contains supplementary material, which is available to authorized users.

What Next for Quantum Mechanics in Structure-Based Drug Discovery?

There is significant potential for electronic structure methods to improve the quality of the predictions furnished by the tools of computer-aided drug design, which typically rely on empirically derived functions. In this perspective, we consider some recent examples of how quantum mechanics has been applied in predicting protein-ligand geometries, protein-ligand binding affinities and ligand strain on binding. We then outline several significant developments in quantum mechanics methodology likely to influence these approaches: in particular, we note the advent of more computationally expedient ab initio quantum mechanical methods that can provide chemical accuracy for larger molecular systems than hitherto possible. We highlight the emergence of increasingly accurate semiempirical quantum mechanical methods and the associated role of machine learning and molecular databases in their development. Indeed, the convergence of improved algorithms for solving and analyzing electronic structure, modern machine learning methods, and increasingly comprehensive benchmark data sets of molecular geometries and energies provides a context in which the potential of quantum mechanics will be increasingly realized in driving future developments and applications in structure-based drug discovery.

Conformational Searching with Quantum Mechanics

Estimating the range of three-dimensional structures (conformations) that are available to a molecule is a key component of computer-aided drug design. Quantum mechanical simulation offers improved accuracy over forcefield methods, but at a high computational cost. The question is whether this increased cost can be justified in a context in which high-throughput analysis of large numbers of molecules is often key. This chapter discusses the application of quantum mechanics to conformational searching, with a focus on three key challenges: (1) the generation of ensembles that include a good approximation to a molecule's bioactive conformation at as prominent a ranking as possible; (2) rational analysis and modification of a pre-established bioactive conformation in terms of its energetics; and (3) approximation of real solution-phase conformational ensembles in tandem with NMR data. The impact of QM on the high-throughput application (1) is debatable, meaning that for the moment its primary application is still lower-throughput applications such as (2) and (3). The optimal choice of QM method is also discussed. Rigorous benchmarking suggests that DFT methods are only acceptable when used with large basis sets, but a trickle of papers continue to obtain useful results with relatively low-cost methods, leading to a dilemma that the literature has yet to fully resolve.

Fast, efficient fragment-based coordinate generation for Open Babel

Rapidly predicting an accurate three dimensional geometry of a molecule is a crucial task for cheminformatics and across a wide range of molecular modeling. Consequently, developing a fast, accurate, and open implementation of structure prediction is necessary for reproducible cheminformatics research. We introduce a fragment-based coordinate generation implementation for Open Babel, a widely-used open source toolkit for cheminformatics. The new implementation improves speed and stereochemical accuracy, while retaining or improving accuracy of bond lengths, bond angles, and dihedral torsions. Input molecules are broken into fragments by cutting at rotatable bonds. The coordinates of fragments are set according to a fragment library, prepared from open crystallographic databases. Since the coordinates of multiple atoms are decided at once, coordinate prediction is accelerated over the previous rules-based implementation in Open Babel, as well as the widely-used distance geometry methods in RDKit. This new implementation will be beneficial for a wide range of applications, including computational property prediction in polymers, molecular materials and drug design.

The upcoming subatomic resolution revolution

Subatomic resolution macromolecular crystallography has been revealing the most fascinating details of macromolecular structures for many years. This most extreme form of macromolecular crystallography is going through rapid changes. A new generation of superbrilliant X-ray sources and detectors is facilitating the rapid acquisition of high-quality datasets. Equally important, a new breed of methods and highly integrated advanced computational tools for structure refinement and analysis is poised to change the way we use subatomic resolution data and reposition high-resolution macromolecular crystallography in medicinal chemistry studies. Subatomic resolution macromolecular crystallography may soon be a routine source of detailed molecular information besides precise geometries, including binding energies and other chemical descriptors, opening new possibilities of application.

Energy windows for computed compound conformers: covering artefacts or truly large reorganization energies?

The generation of 3D conformers of small molecules underpins most computational drug discovery. Thus, the conformer quality is critical and depends on their energetics. A key parameter is the empirical conformational energy window (ΔEw), since only conformers within ΔEw are retained. However, ΔEw values in use appear unrealistically large. We analyze the factors pertaining to the conformer energetics and ΔEw. We argue that more attention must be focused on the problem of collapsed low-energy conformers. That is due to artificial intramolecular stabilization and occurs even with continuum solvation. Consequently, the conformational energy of extended bioactive structures is artefactually increased, which inflates ΔEw. Thus, this Perspective highlights the issues arising from low-energy conformers and suggests improvements via empirical or physics-based strategies.