Revolutionizing GPCR-ligand predictions: DeepGPCR with experimental validation for high-precision drug discovery.

G-protein coupled receptors (GPCRs), crucial in various diseases, are targeted of over 40% of approved drugs. However, the reliable acquisition of experimental GPCRs structures is hindered by their lipid-embedded conformations. Traditional protein-ligand interaction models falter in GPCR-drug interactions, caused by limited and low-quality structures. Generalized models, trained on soluble protein-ligand pairs, are also inadequate. To address these issues, we developed two models, DeepGPCR_BC for binary classification and DeepGPCR_RG for affinity prediction. These models use non-structural GPCR-ligand interaction data, leveraging graph convolutional networks and mol2vec techniques to represent binding pockets and ligands as graphs. This approach significantly speeds up predictions while preserving critical physical-chemical and spatial information. In independent tests, DeepGPCR_BC surpassed Autodock Vina and Schrödinger Dock with an area under the curve of 0.72, accuracy of 0.68 and true positive rate of 0.73, whereas DeepGPCR_RG demonstrated a Pearson correlation of 0.39 and root mean squared error of 1.34. We applied these models to screen drug candidates for GPR35 (Q9HC97), yielding promising results with three (F545-1970, K297-0698, S948-0241) out of eight candidates. Furthermore, we also successfully obtained six active inhibitors for GLP-1R. Our GPCR-specific models pave the way for efficient and accurate large-scale virtual screening, potentially revolutionizing drug discovery in the GPCR field.

Deep Learning-Based construction of a Drug-Like compound database and its application in virtual screening of HsDHODH inhibitors.

The process of virtual screening relies heavily on the databases, but it is disadvantageous to conduct virtual screening based on commercial databases with patent-protected compounds, high compound toxicity and side effects. Therefore, this paper utilizes generative recurrent neural networks (RNN) containing long short-term memory (LSTM) cells to learn the properties of drug compounds in the DrugBank, aiming to obtain a new and virtual screening compounds database with drug-like properties. Ultimately, a compounds database consisting of 26,316 compounds is obtained by this method. To evaluate the potential of this compounds database, a series of tests are performed, including chemical space, ADME properties, compound fragmentation, and synthesizability analysis. As a result, it is proved that the database is equipped with good drug-like properties and a relatively new backbone, its potential in virtual screening is further tested. Finally, a series of seedling compounds with completely new backbones are obtained through docking and binding free energy calculations.

A deep learning based multi-model approach for predicting drug-like chemical compound's toxicity.

Ensuring the safety and efficacy of chemical compounds is crucial in small-molecule drug development. In the later stages of drug development, toxic compounds pose a significant challenge, losing valuable resources and time. Early and accurate prediction of compound toxicity using deep learning models offers a promising solution to mitigate these risks during drug discovery. In this study, we present the development of several deep-learning models aimed at evaluating different types of compound toxicity, including acute toxicity, carcinogenicity, hERG_cardiotoxicity (the human ether-a-go-go related gene caused cardiotoxicity), hepatotoxicity, and mutagenicity. To address the inherent variations in data size, label type, and distribution across different types of toxicity, we employed diverse training strategies. Our first approach involved utilizing a graph convolutional network (GCN) regression model to predict acute toxicity, which achieved notable performance with Pearson R 0.76, 0.74, and 0.65 for intraperitoneal, intravenous, and oral administration routes, respectively. Furthermore, we trained multiple GCN binary classification models, each tailored to a specific type of toxicity. These models exhibited high area under the curve (AUC) scores, with an impressive AUC of 0.69, 0.77, 0.88, and 0.79 for predicting carcinogenicity, hERG_cardiotoxicity, mutagenicity, and hepatotoxicity, respectively. Additionally, we have used the approved drug dataset to determine the appropriate threshold value for the prediction score in model usage. We integrated these models into a virtual screening pipeline to assess their effectiveness in identifying potential low-toxicity drug candidates. Our findings indicate that this deep learning approach has the potential to significantly reduce the cost and risk associated with drug development by expediting the selection of compounds with low toxicity profiles. Therefore, the models developed in this study hold promise as critical tools for early drug candidate screening and selection.

Generating and screening de novo compounds against given targets using ultrafast deep learning models as core components.

Deep learning is an artificial intelligence technique in which models express geometric transformations over multiple levels. This method has shown great promise in various fields, including drug development. The availability of public structure databases prompted the researchers to use generative artificial intelligence models to narrow down their search of the chemical space, a novel approach to chemogenomics and de novo drug development. In this study, we developed a strategy that combined an accelerated LSTM_Chem (long short-term memory for de novo compounds generation), dense fully convolutional neural network (DFCNN), and docking to generate a large number of de novo small molecular chemical compounds for given targets. To demonstrate its efficacy and applicability, six important targets that account for various human disorders were used as test examples. Moreover, using the M protease as a proof-of-concept example, we find that iteratively training with previously selected candidates can significantly increase the chance of obtaining novel compounds with higher and higher predicted binding affinities. In addition, we also check the potential benefit of obtaining reliable final de novo compounds with the help of MD simulation and metadynamics simulation. The generation of de novo compounds and the discovery of binders against various targets proposed here would be a practical and effective approach. Assessing the efficacy of these top de novo compounds with biochemical studies is promising to promote related drug development.

Substituent effects on the selectivity of ambimodal [6+4]/[4+2] cycloaddition.

In this work, we report a density functional theory (DFT) study and a dynamical trajectory study of substituent effects on the ambimodal [6+4]/[4+2] cycloaddition proposed for 1,3,5,10,12-cycloheptadecapentaene, referred to as cycloheptadecapentaene. The proposed cycloaddition proceeds through an ambimodal transition state, which results in both a [6+4] adduct a [4+2] adduct directly. The [6+4] adduct can be readily converted to the [4+2] adduct via a Cope rearrangement. We study the selectivity of the reaction with regard to the position of substituents, steric effects of substituents, and electronic effects of substituents. In the dynamical trajectory study, we find that nitro-substituted reactants lead to a new product from the ambimodal transition state via the hetero Diels-Alder reaction, and this new product can then be converted to a [4+2] adduct by a hetero [3, 3]-sigmatropic rearrangement. These results may provide insights for designing more bridged heterocyclic compounds.

Biosynthesis of eriodictyol in citrus waster by endowing P450BM3 activity of naringenin hydroxylation.

The flavonoid naringenin is abundantly present in pomelo peels, and the unprocessed naringenin in wastes is not friendly for the environment once discarded directly. Fortunately, the hydroxylated product of eriodictyol from naringenin exhibits remarkable antioxidant and anticancer properties. The P450s was suggested promising for the bioconversion of the flavonoids, but less naturally existed P450s show hydroxylation activity to C3' of the naringenin. By well analyzing the catalytic mechanism and the conformations of the naringenin in P450, we proposed that the intermediate Cmpd I ((porphyrin)Fe = O) is more reasonable as key conformation for the hydrolyzation, and the distance between C3'/C5' of naringenin to the O atom of CmpdI determines the hydroxylating activity for the naringenin. Thus, the "flying kite model" that gradually drags the C-H bond of the substrate to the O atom of CmpdI was put forward for rational design. With ab initio design, we successfully endowed the self-sufficient P450-BM3 hydroxylic activity to naringenin and obtained mutant M5-5, with kcat, Km, and kcat/Km values of 230.45 min-1, 310.48 µM, and 0.742 min-1 µM-1, respectively. Furthermore, the mutant M4186 was screened with kcat/Km of 4.28-fold highly improved than the reported M13. The M4186 also exhibited 62.57% yield of eriodictyol, more suitable for the industrial application. This study provided a theoretical guide for the rational design of P450s to the nonnative compounds. KEY POINTS: •The compound I is proposed as the starting point for the rational design of the P450BM3 •"Flying kite model" is proposed based on the distance between O of Cmpd I and C3'/C5' of naringenin •Mutant M15-5 with 1.6-fold of activity than M13 was obtained by ab initio modification.

Entropy driven cooperativity effect in multi-site drug optimization targeting SARS-CoV-2 papain-like protease.

Papain-like protease (PLpro), a non-structural protein encoded by SARS-CoV-2, is an important therapeutic target. Regions 1 and 5 of an existing drug, GRL0617, can be optimized to produce cooperativity with PLpro binding, resulting in stronger binding affinity. This work investigated the origin of the cooperativity using molecular dynamics simulations combined with the interaction entropy (IE) method. The regions' improvement exhibits cooperativity by calculating the binding free energies between the complex of PLpro-inhibitor. The thermodynamic integration method further verified the cooperativity generated in the drug improvement. To further determine the specific source of cooperativity, enthalpy and entropy in the complexes were calculated using molecular mechanics/generalized Born surface area and IE. The results show that the entropic change is an important contributor to the cooperativity. Our study also identified residues P248, Q269, and T301 that play a significant role in cooperativity. The optimization of the inhibitor stabilizes these residues and minimizes the entropic loss, and the cooperativity observed in the binding free energy can be attributed to the change in the entropic contribution of these residues. Based on our research, the application of cooperativity can facilitate drug optimization, and provide theoretical ideas for drug development that leverage cooperativity by reducing the contribution of entropy through multi-locus binding.

An Efficient Approach to the Accurate Prediction of Mutational Effects in Antigen Binding to the MHC1.

The major histocompatibility complex (MHC) can recognize and bind to external peptides to generate effective immune responses by presenting the peptides to T cells. Therefore, understanding the binding modes of peptide-MHC complexes (pMHC) and predicting the binding affinity of pMHCs play a crucial role in the rational design of peptide vaccines. In this study, we employed molecular dynamics (MD) simulations and free energy calculations with an Alanine Scanning with Generalized Born and Interaction Entropy (ASGBIE) method to investigate the protein-peptide interaction between HLA-A*02:01 and the G9209 peptide derived from the melanoma antigen gp100. The energy contribution of individual residue was calculated using alanine scanning, and hotspots on both the MHC and the peptides were identified. Our study shows that the pMHC binding is dominated by the van der Waals interactions. Furthermore, we optimized the ASGBIE method, achieving a Pearson correlation coefficient of 0.91 between predicted and experimental binding affinity for mutated antigens. This represents a significant improvement over the conventional MM/GBSA method, which yields a Pearson correlation coefficient of 0.22. The computational protocol developed in this study can be applied to the computational screening of antigens for the MHC1 as well as other protein-peptide binding systems.

Rational Design of a Highly Specific Prolyl Endopeptidase To Activate the Antihypertensive Effect of Peptides.

Enzymatic hydrolysis of food-derived proteins to produce bioactive peptides could activate food functions such as antihypertension. However, the diversity of enzymatic hydrolysis products can reduce bioactive peptides' efficacy. Highly specific proteases can homogenize the hydrolysis products to reduce the production of impotent peptides. In this study, we successfully obtained M. xanthus prolyl endopeptidase mutant Y451M by constraint/free molecular dynamics simulations and binding energy calculations. The specificity of Y451M for proline was increased by 286 % compared to WT, while its activity was almost unchanged. Milk-derived substrates processed with Y451M showed an antihypertensive effect that was 567 % higher than without enzymes. The ability to activate food antihypertension increased 152 % and the use of enzyme by 192 % compared with WT. Specific proteases are thus valuable tools in the processing of complex substrates to obtain bioactive peptides.

Combined Antibodies Evusheld against the SARS-CoV-2 Omicron Variants BA.1.1 and BA.5: Immune Escape Mechanism from Molecular Simulation.

The Omicron lineage of SARS-CoV-2, which was first reported in November 2021, has spread globally and become dominant, splitting into several sublineages. Experiments have shown that Omicron lineage has escaped or reduced the activity of existing monoclonal antibodies, but the origin of escape mechanism caused by mutation is still unknown. This work uses molecular dynamics and umbrella sampling methods to reveal the escape mechanism of BA.1.1 to monoclonal antibody (mAb) Tixagevimab (AZD1061) and BA.5 to mAb Cilgavimab (AZD8895), both mAbs were combined to form antibody cocktail, Evusheld (AZD7442). The binding free energy of BA.1.1-AZD1061 and BA.5-AZD8895 has been severely reduced due to multiple-site mutated Omicron variants. Our results show that the two Omicron variants, which introduce a substantial number of positively charged residues, can weaken the electrostatic attraction between the receptor binding domain (RBD) and AZD7442, thus leading to a decrease in affinity. Additionally, using umbrella sampling along dissociation pathway, we found that the two Omicron variants severely impaired the interaction between the RBD of SARS-CoV-2's spike glycoprotein (S protein) and complementary determining regions (CDRs) of mAbs, especially in CDR3H. Although mAbs AZD8895 and AZD1061 are knocked out by BA.5 and BA.1.1, respectively, our results confirm that the antibody cocktail AZD7442 retains activity against BA.1.1 and BA.5 because another antibody is still on guard. The study provides theoretical insights for mAbs interacting with BA.1.1 and BA.5 from both energetic and dynamic perspectives, and we hope this will help in developing new monoclonals and combinations to protect those unable to mount adequate vaccine responses.

MolTaut: A Tool for the Rapid Generation of Favorable Tautomer in Aqueous Solution.

Fast and proper treatment of the tautomeric states for drug-like molecules is critical in computer-aided drug discovery since the major tautomer of a molecule determines its pharmacophore features and physical properties. We present MolTaut, a tool for the rapid generation of favorable states of drug-like molecules in water. MolTaut works by enumerating possible tautomeric states with tautomeric transformation rules, ranking tautomers with their relative internal energies and solvation energies calculated by AI-based models, and generating preferred ionization states according to predicted microscopic pKa. Our test shows that the ranking ability of the AI-based tautomer scoring approach is comparable to the DFT method (wB97X/6-31G*//M062X/6-31G*/SMD) from which the AI models try to learn. We find that the substitution effect on tautomeric equilibrium is well predicted by MolTaut, which is helpful in computer-aided ligand design. The source code of MolTaut is freely available to researchers and can be accessed at https://github.com/xundrug/moltaut. To facilitate the usage of MolTaut by medicinal chemists, we made a free web server, which is available at http://moltaut.xundrug.cn. MolTaut is a handy tool for investigating the tautomerization issue in drug discovery.

Screening Power of End-Point Free-Energy Calculations in Cucurbituril Host-Guest Systems.

End-point free-energy methods as an indispensable component in virtual screening are commonly recognized as a tool with a certain level of screening power in pharmaceutical research. While a huge number of records could be found for end-point applications in protein-ligand, protein-protein, and protein-DNA complexes from academic and industrial reports, up to now, there is no large-scale benchmark in host-guest complexes supporting the screening power of end-point free-energy techniques. A good benchmark requires a data set of sufficient coverage of pharmaceutically relevant chemical space, a long-time sampling length supporting the trajectory approximation of the ensemble average, and a sufficient sample size of receptor-acceptor pairs to stabilize the performance statistics. In this work, selecting a popular family of macrocyclic hosts named cucurbiturils, we construct a large data set containing 154 host-guest pairs, perform extensive end-point sampling of several hundred nanosecond lengths for each system, and extract the free-energy estimates with a variety of end-point free-energy techniques, including the advanced three-trajectory dielectric-constant-variable regime proposed in our recent work. The best-performing end-point protocol employs GAFF2 for solute descriptions, the three-trajectory end-point sampling regime, and the MM/GBSA Hamiltonian in free-energy extraction, achieving a high ranking metrics of Kendall τ > 0.6, a Pearlman predictive index of ∼0.8, and a high scoring power of Pearson r > 0.8. The current project as the first large-scale systematic benchmark of end-point methods in host-guest complexes in academic publications provides solid evidence of the applicability of end-point techniques and direct guidance of computational setups in practical host-guest systems.

Deep Learning-Based Bioactive Therapeutic Peptide Generation and Screening.

Many bioactive peptides demonstrated therapeutic effects over complicated diseases, such as antiviral, antibacterial, anticancer, etc. It is possible to generate a large number of potentially bioactive peptides using deep learning in a manner analogous to the generation of de novo chemical compounds using the acquired bioactive peptides as a training set. Such generative techniques would be significant for drug development since peptides are much easier and cheaper to synthesize than compounds. Despite the limited availability of deep learning-based peptide-generating models, we have built an LSTM model (called LSTM_Pep) to generate de novo peptides and fine-tuned the model to generate de novo peptides with specific prospective therapeutic benefits. Remarkably, the Antimicrobial Peptide Database has been effectively utilized to generate various kinds of potential active de novo peptides. We proposed a pipeline for screening those generated peptides for a given target and used the main protease of SARS-COV-2 as a proof-of-concept. Moreover, we have developed a deep learning-based protein-peptide prediction model (DeepPep) for rapid screening of the generated peptides for the given targets. Together with the generating model, we have demonstrated that iteratively fine-tuning training, generating, and screening peptides for higher-predicted binding affinity peptides can be achieved. Our work sheds light on developing deep learning-based methods and pipelines to effectively generate and obtain bioactive peptides with a specific therapeutic effect and showcases how artificial intelligence can help discover de novo bioactive peptides that can bind to a particular target.

Binding of berberine derivates to G-quadruplex: insight from a computational study.

Human telomerase exhibits significant activity in cancer cells relative to normal cells, which contributes to the immortal proliferation of cancer cells. To counter this, the stabilization of G-quadruplexes formed in the guanine-rich sequence of the cancer cell chromosome has emerged as a promising avenue for anti-cancer therapy. Berberine (BER), an alkaloid that is derived from traditional Chinese medicines, has shown potential for stabilizing G-quadruplexes. To investigate the atomic interactions between G-quadruplexes and BER and its derivatives, molecular dynamics simulations were conducted. Modeling the interactions between G-quadruplexes and ligands accurately is challenging due to the strong negative charge of nucleic acids. Thus, various force fields and charge models for the G-quadruplex and ligands were tested to obtain precise simulation results. The binding energies were calculated by a combination of molecular mechanics/generalized Born surface area and interaction entropy methods, and the calculated results correlated well with experimental results. B-factor and hydrogen bond analyses demonstrated that the G-quadruplex was more stable in the presence of ligands than in the absence of ligands. Calculation of the binding free energy showed that the BER derivatives bind to a G-quadruplex with higher affinity than that of BER. The breakdown of the binding free energy to per-nucleotide energies suggested that the first G-tetrad played a primary role in binding. Additionally, energy and geometric properties analyses indicated that van der Waals interactions were the most favorable interactions between the derivatives and the G-quadruplexes. Overall, these findings provide crucial atomic-level insights into the binding of G-quadruplexes and their inhibitors.

DeepBindGCN: Integrating Molecular Vector Representation with Graph Convolutional Neural Networks for Protein-Ligand Interaction Prediction.

The core of large-scale drug virtual screening is to select the binders accurately and efficiently with high affinity from large libraries of small molecules in which non-binders are usually dominant. The binding affinity is significantly influenced by the protein pocket, ligand spatial information, and residue types/atom types. Here, we used the pocket residues or ligand atoms as the nodes and constructed edges with the neighboring information to comprehensively represent the protein pocket or ligand information. Moreover, the model with pre-trained molecular vectors performed better than the one-hot representation. The main advantage of DeepBindGCN is that it is independent of docking conformation, and concisely keeps the spatial information and physical-chemical features. Using TIPE3 and PD-L1 dimer as proof-of-concept examples, we proposed a screening pipeline integrating DeepBindGCN and other methods to identify strong-binding-affinity compounds. It is the first time a non-complex-dependent model has achieved a root mean square error (RMSE) value of 1.4190 and Pearson r value of 0.7584 in the PDBbind v.2016 core set, respectively, thereby showing a comparable prediction power with the state-of-the-art affinity prediction models that rely upon the 3D complex. DeepBindGCN provides a powerful tool to predict the protein-ligand interaction and can be used in many important large-scale virtual screening application scenarios.

Comprehensive Evaluation of End-Point Free Energy Techniques in Carboxylated-Pillar[6]arene Host-Guest Binding: III. Force-Field Comparison, Three-Trajectory Realization and Further Dielectric Augmentation.

Host-guest binding, despite the relatively simple structural and chemical features of individual components, still poses a challenge in computational modelling. The extreme underperformance of standard end-point methods in host-guest binding makes them practically useless. In the current work, we explore a potentially promising modification of the three-trajectory realization. The alteration couples the binding-induced structural reorganization into free energy estimation and suffers from dramatic fluctuations in internal energies in protein-ligand situations. Fortunately, the relatively small size of host-guest systems minimizes the magnitude of internal fluctuations and makes the three-trajectory realization practically suitable. Due to the incorporation of intra-molecular interactions in free energy estimation, a strong dependence on the force field parameters could be incurred. Thus, a term-specific investigation of transferable GAFF derivatives is presented, and noticeable differences in many aspects are identified between commonly applied GAFF and GAFF2. These force-field differences lead to different dynamic behaviors of the macrocyclic host, which ultimately would influence the end-point sampling and binding thermodynamics. Therefore, the three-trajectory end-point free energy calculations are performed with both GAFF versions. Additionally, due to the noticeable differences between host dynamics under GAFF and GAFF2, we add additional benchmarks of the single-trajectory end-point calculations. When only the ranks of binding affinities are pursued, the three-trajectory realization performs very well, comparable to and even better than the regressed PBSA_E scoring function and the dielectric constant-variable regime. With the GAFF parameter set, the TIP3P water in explicit solvent sampling and either PB or GB implicit solvent model in free energy estimation, the predictive power of the three-trajectory realization in ranking calculations surpasses all existing end-point methods on this dataset. We further combine the three-trajectory realization with another promising modified end-point regime of varying the interior dielectric constant. The combined regime does not incur sizable improvements for ranks and deviations from experiment exhibit non-monotonic variations.

HergSPred: Accurate Classification of hERG Blockers/Nonblockers with Machine-Learning Models.

The human ether-à-go-go-related gene (hERG) K+ channel plays an important role in cardiac action potentials. The inhibition of the hERG channel may lead to long QT syndrome (LQTS) and even sudden cardiac death. Due to severe hERG-related cardiotoxicity, many drugs have been withdrawn from the market. Therefore, it is necessary to estimate the chemical blockade of hERG in the early stage of drug discovery. In this study, we collected 12,850 compounds with hERG inhibition data from the literature and trained a series of hERG blocking classification models based on the MACCS and Morgan fingerprints. A consensus model named HergSPred was generated based on the individual models using voting principles. The accuracy of HergSPred is higher than previous models using identical training and test sets. Moreover, we analyzed the contribution of each input fingerprint to the prediction output to obtain intuitive chemical insights into the hERG inhibition, which allows visualization of warning substructures that may cause cardiotoxicity in the input compound. The model is available at http://www.icdrug.com/ICDrug/T.

AA-Score: a New Scoring Function Based on Amino Acid-Specific Interaction for Molecular Docking.

The protein-ligand scoring function plays an important role in computer-aided drug discovery and is heavily used in virtual screening and lead optimization. In this study, we developed a new empirical protein-ligand scoring function with amino acid-specific interaction components for hydrogen bond, van der Waals, and electrostatic interactions. In addition, hydrophobic, π-stacking, π-cation, and metal-ligand interactions are also included in the new scoring function. To better evaluate the performance of the AA-Score, we generated several new test sets for evaluation of scoring, ranking, and docking performances, respectively. Extensive tests show that AA-Score performs well on scoring, docking, and ranking as compared to other widely used traditional scoring functions. The performance improvement of AA-Score benefits from the decomposition of individual interaction into amino acid-specific types. To facilitate applications, we developed an easy-to-use tool to analyze protein-ligand interaction fingerprint and predict binding affinity using the AA-Score. The source code and associated running examples can be found at https://github.com/xundrug/AA-Score-Tool.

Comprehensive evaluation of end-point free energy techniques in carboxylated-pillar[6]arene host-guest binding: I. Standard procedure.

Despite the massive application of end-point free energy methods in protein-ligand and protein-protein interactions, computational understandings about their performance in relatively simple and prototypical host-guest systems are limited. In this work, we present a comprehensive benchmark calculation with standard end-point free energy techniques in a recent host-guest dataset containing 13 host-guest pairs involving the carboxylated-pillar[6]arene host. We first assess the charge schemes for solutes by comparing the charge-produced electrostatics with many ab initio references, in order to obtain a preliminary albeit detailed view of the charge quality. Then, we focus on four modelling details of end-point free energy calculations, including the docking procedure for the generation of initial condition, the charge scheme for host and guest molecules, the water model used in explicit-solvent sampling, and the end-point methods for free energy estimation. The binding thermodynamics obtained with different modelling schemes are compared with experimental references, and some practical guidelines on maximizing the performance of end-point methods in practical host-guest systems are summarized. Further, we compare our simulation outcome with predictions in the grand challenge and discuss further developments to improve the prediction quality of end-point free energy methods. Overall, unlike the widely acknowledged applicability in protein-ligand binding, the standard end-point calculations cannot produce useful outcomes in host-guest binding and thus are not recommended unless alterations are performed.

Comprehensive evaluation of end-point free energy techniques in carboxylated-pillar[6]arene host-guest binding: II. regression and dielectric constant.

End-point free energy calculations as a powerful tool have been widely applied in protein-ligand and protein-protein interactions. It is often recognized that these end-point techniques serve as an option of intermediate accuracy and computational cost compared with more rigorous statistical mechanic models (e.g., alchemical transformation) and coarser molecular docking. However, it is observed that this intermediate level of accuracy does not hold in relatively simple and prototypical host-guest systems. Specifically, in our previous work investigating a set of carboxylated-pillar[6]arene host-guest complexes, end-point methods provide free energy estimates deviating significantly from the experimental reference, and the rank of binding affinities is also incorrectly computed. These observations suggest the unsuitability and inapplicability of standard end-point free energy techniques in host-guest systems, and alteration and development are required to make them practically usable. In this work, we consider two ways to improve the performance of end-point techniques. The first one is the PBSA_E regression that varies the weights of different free energy terms in the end-point calculation procedure, while the second one is considering the interior dielectric constant as an additional variable in the end-point equation. By detailed investigation of the calculation procedure and the simulation outcome, we prove that these two treatments (i.e., regression and dielectric constant) are manipulating the end-point equation in a somehow similar way, i.e., weakening the electrostatic contribution and strengthening the non-polar terms, although there are still many detailed differences between these two methods. With the trained end-point scheme, the RMSE of the computed affinities is improved from the standard ~ 12 kcal/mol to ~ 2.4 kcal/mol, which is comparable to another altered end-point method (ELIE) trained with system-specific data. By tuning PBSA_E weighting factors with the host-specific data, it is possible to further decrease the prediction error to ~ 2.1 kcal/mol. These observations along with the extremely efficient optimized-structure computation procedure suggest the regression (i.e., PBSA_E as well as its GBSA_E extension) as a practically applicable solution that brings end-point methods back into the library of usable tools for host-guest binding. However, the dielectric-constant-variable scheme cannot effectively minimize the experiment-calculation discrepancy for absolute binding affinities, but is able to improve the calculation of affinity ranks. This phenomenon is somehow different from the protein-ligand case and suggests the difference between host-guest and biomacromolecular (protein-ligand and protein-protein) systems. Therefore, the spectrum of tools usable for protein-ligand complexes could be unsuitable for host-guest binding, and numerical validations are necessary to screen out really workable solutions in these 'prototypical' situations.