The Impact of Supervised Learning Methods in Ultralarge High-Throughput Docking.

Structure-based virtual screening methods are, nowadays, one of the key pillars of computational drug discovery. In recent years, a series of studies have reported docking-based virtual screening campaigns of large databases ranging from hundreds to thousands of millions compounds, further identifying novel hits after experimental validation. As these larg-scale efforts are not generally accessible, machine learning-based protocols have emerged to accelerate the identification of virtual hits within an ultralarge chemical space, reaching impressive reductions in computational time. Herein, we illustrate the motivation and the problem behind the screening of large databases, providing an overview of key concepts and essential applications of machine learning-accelerated protocols, specifically concerning supervised learning methods. We also discuss where the field stands with these novel developments, highlighting possible insights for future studies.

How good are AlphaFold models for docking-based virtual screening?

A crucial component in structure-based drug discovery is the availability of high-quality three-dimensional structures of the protein target. Whenever experimental structures were not available, homology modeling has been, so far, the method of choice. Recently, AlphaFold (AF), an artificial-intelligence-based protein structure prediction method, has shown impressive results in terms of model accuracy. This outstanding success prompted us to evaluate how accurate AF models are from the perspective of docking-based drug discovery. We compared the high-throughput docking (HTD) performance of AF models with their corresponding experimental PDB structures using a benchmark set of 22 targets. The AF models showed consistently worse performance using four docking programs and two consensus techniques. Although AlphaFold shows a remarkable ability to predict protein architecture, this might not be enough to guarantee that AF models can be reliably used for HTD, and post-modeling refinement strategies might be key to increase the chances of success.

Solvent effects on the NMR shieldings of stacked DNA base pairs.

Stacking effects are among the most important effects in DNA. We have recently studied their influence in fragments of DNA through the analysis of NMR magnetic shieldings, firstly in vacuo. As a continuation of this line of research we show here the influence of solvent effects on the shieldings through the application of both explicit and implicit models. We found that the explicit solvent model is more appropriate for consideration due to the results matching better in general with experiments, as well as providing clear knowledge of the electronic origin of the value of the shieldings. Our study is grounded on a recently developed theoretical model of our own, by which we are able to learn about the magnetic effects of given fragments of DNA molecules on selected base pairs. We use the shieldings of the atoms of a central base pair (guanine-cytosine) of a selected fragment of DNA molecules as descriptors of physical effects, like π-stacking and solvent effects. They can be taken separately and altogether. The effect of π-stacking is introduced through the addition of some pairs above and below of the central base pair, and now, the solvent effect is considered including a network of water molecules that consist of two solvation layers, which were fixed in the calculations performed in all fragments. We show that the solvent effects enhance the stacking effects on the magnetic shieldings of atoms that belong to the external N-H bonds. The net effect is of deshielding on both atoms. There is also a deshielding effect on the carbon atoms that belong to CO bonds, for which the oxygen atom has an explicit hydrogen bond (HB) with a solvent water molecule. Solvent effects are found to be no higher than a few percent of the total value of the shieldings (between 1% and 5%) for most atoms, although there are few for which such an effect can be higher. There is one nitrogen atom, the acceptor of the HB between guanine and cytosine, that is more highly shielded (around 15 ppm or 10%) when the explicit solvent is considered. In a similar manner, the most external nitrogen atom of cytosine and the hydrogen atom that is bonded to it are highly deshielded (around 10 ppm for nitrogen and around 3 ppm for hydrogen).

Insights into the product release mechanism of dengue virus NS3 helicase.

The non-structural protein 3 helicase (NS3h) is a multifunctional protein that is critical in RNA replication and other stages in the flavivirus life cycle. NS3h uses energy from ATP hydrolysis to translocate along single stranded nucleic acid and to unwind double stranded RNA. Here we present a detailed mechanistic analysis of the product release stage in the catalytic cycle of the dengue virus (DENV) NS3h. This study is based on a combined experimental and computational approach of product-inhibition studies and free energy calculations. Our results support a model in which the catalytic cycle of ATP hydrolysis proceeds through an ordered sequential mechanism that includes a ternary complex intermediate (NS3h-Pi-ADP), which evolves releasing the first product, phosphate (Pi), and subsequently ADP. Our results indicate that in the product release stage of the DENV NS3h a novel open-loop conformation plays an important role that may be conserved in NS3 proteins of other flaviviruses as well.

Guided structure-based ligand identification and design via artificial intelligence modeling.

INTRODUCTION: The implementation of Artificial Intelligence (AI) methodologies to drug discovery (DD) are on the rise. Several applications have been developed for structure-based DD, where AI methods provide an alternative framework for the identification of ligands for validated therapeutic targets, as well as the de novo design of ligands through generative models. AREAS COVERED: Herein, the authors review the contributions between the 2019 to present period regarding the application of AI methods to structure-based virtual screening (SBVS) which encompasses mainly molecular docking applications - binding pose prediction and binary classification for ligand or hit identification-, as well as de novo drug design driven by machine learning (ML) generative models, and the validation of AI models in structure-based screening. Studies are reviewed in terms of their main objective, used databases, implemented methodology, input and output, and key results . EXPERT OPINION: More profound analyses regarding the validity and applicability of AI methods in DD have begun to appear. In the near future, we expect to see more structure-based generative models- which are scarce in comparison to ligand-based generative models-, the implementation of standard guidelines for validating the generated structures, and more analyses regarding the validation of AI methods in structure-based DD.

Machine Learning Toxicity Prediction: Latest Advances by Toxicity End Point.

Machine learning (ML) models to predict the toxicity of small molecules have garnered great attention and have become widely used in recent years. Computational toxicity prediction is particularly advantageous in the early stages of drug discovery in order to filter out molecules with high probability of failing in clinical trials. This has been helped by the increase in the number of large toxicology databases available. However, being an area of recent application, a greater understanding of the scope and applicability of ML methods is still necessary. There are various kinds of toxic end points that have been predicted in silico. Acute oral toxicity, hepatotoxicity, cardiotoxicity, mutagenicity, and the 12 Tox21 data end points are among the most commonly investigated. Machine learning methods exhibit different performances on different data sets due to dissimilar complexity, class distributions, or chemical space covered, which makes it hard to compare the performance of algorithms over different toxic end points. The general pipeline to predict toxicity using ML has already been analyzed in various reviews. In this contribution, we focus on the recent progress in the area and the outstanding challenges, making a detailed description of the state-of-the-art models implemented for each toxic end point. The type of molecular representation, the algorithm, and the evaluation metric used in each research work are explained and analyzed. A detailed description of end points that are usually predicted, their clinical relevance, the available databases, and the challenges they bring to the field are also highlighted.

Evaluation of Silicone Fluids and Resins as CO2 Thickeners for Enhanced Oil Recovery Using a Computational and Experimental Approach.

CO2 thickeners have the potential to be a game changer for enhanced oil recovery, carbon capture utilization and storage, and hydraulic fracturing. Thickener design is challenging due to polymers' low solubility in supercritical CO2 (scCO2) and the difficulty of substantially increasing the viscosity of CO2. In this contribution, we present a framework to design CO2 soluble thickeners, combining calculations using a quantum mechanical solvation model with direct laboratory viscosity testing. The conductor-like polarizable continuum model for solvation free-energy calculations was used to determine functional silicone and silsesquioxane solubilities in scCO2. This method allowed for a fast and efficient identification of CO2-soluble compounds, revealing silsesquioxanes as more CO2-philic than linear polydimethylsiloxane (PDMS), the most efficient non-fluorinated thickener know to date. The rolling ball apparatus was used to measure the viscosity of scCO2 with both PDMS and silicone resins with added silica nanoparticles. Methyl silicone resins were found to be stable and fast to disperse in scCO2 while having a significant thickening effect. They have a larger effect on the solution viscosity than higher-molecular-weight PDMS and are able to thicken CO2 even at high temperatures. Silicone resins are thus shown to be promising scCO2 thickeners, exhibiting enhanced solubility and good rheological properties, while also having a moderate cost and being easily commercially attainable.

A Machine Learning Model to Predict Drug Transfer Across the Human Placenta Barrier.

The development of computational models for assessing the transfer of chemicals across the placental membrane would be of the utmost importance in drug discovery campaigns, in order to develop safe therapeutic options. We have developed a low-dimensional machine learning model capable of classifying compounds according to whether they can cross or not the placental barrier. To this aim, we compiled a database of 248 compounds with experimental information about their placental transfer, characterizing each compound with a set of ∼5.4 thousand descriptors, including physicochemical properties and structural features. We evaluated different machine learning classifiers and implemented a genetic algorithm, in a five cross validation scheme, to perform feature selection. The optimization was guided towards models displaying a low number of false positives (molecules that actually cross the placental barrier, but are predicted as not crossing it). A Linear Discriminant Analysis model trained with only four structural features resulted to be robust for this task, exhibiting only one false positive case across all testing folds. This model is expected to be useful in predicting placental drug transfer during pregnancy, and thus could be used as a filter for chemical libraries in virtual screening campaigns.

Artificial intelligence in the early stages of drug discovery.

Although the use of computational methods within the pharmaceutical industry is well established, there is an urgent need for new approaches that can improve and optimize the pipeline of drug discovery and development. In spite of the fact that there is no unique solution for this need for innovation, there has recently been a strong interest in the use of Artificial Intelligence for this purpose. As a matter of fact, not only there have been major contributions from the scientific community in this respect, but there has also been a growing partnership between the pharmaceutical industry and Artificial Intelligence companies. Beyond these contributions and efforts there is an underlying question, which we intend to discuss in this review: can the intrinsic difficulties within the drug discovery process be overcome with the implementation of Artificial Intelligence? While this is an open question, in this work we will focus on the advantages that these algorithms provide over the traditional methods in the context of early drug discovery.

In silico Drug Repurposing for COVID-19: Targeting SARS-CoV-2 Proteins through Docking and Consensus Ranking.

In December 2019, an infectious disease caused by the coronavirus SARS-CoV-2 appeared in Wuhan, China. This disease (COVID-19) spread rapidly worldwide, and on March 2020 was declared a pandemic by the World Health Organization (WHO). Today, over 21 million people have been infected, with more than 750.000 casualties. Today, no vaccine or antiviral drug is available. While the development of a vaccine might take at least a year, and for a novel drug, even longer; finding a new use to an old drug (drug repurposing) could be the most effective strategy. We present a docking-based screening using a quantum mechanical scoring of a library built from approved drugs and compounds undergoing clinical trials, against three SARS-CoV-2 target proteins: the spike or S-protein, and two proteases, the main protease and the papain-like protease. The S-protein binds directly to the Angiotensin Converting Enzyme 2 receptor of the human host cell surface, while the two proteases process viral polyproteins. Following the analysis of our structure-based compound screening, we propose several structurally diverse compounds (either FDA-approved or in clinical trials) that could display antiviral activity against SARS-CoV-2. Clearly, these compounds should be further evaluated in experimental assays and clinical trials to confirm their actual activity against the disease. We hope that these findings may contribute to the rational drug design against COVID-19.

Combination of pose and rank consensus in docking-based virtual screening: the best of both worlds.

The use of high-throughput docking (HTD) in the drug discovery pipeline is today widely established. In spite of methodological improvements in docking accuracy (pose prediction), scoring power, ranking power, and screening power in HTD remain challenging. In fact, pose prediction is of critical importance in view of the pose-dependent scoring process, since incorrect poses will necessarily decrease the ranking power of scoring functions. The combination of results from different docking programs (consensus scoring) has been shown to improve the performance of HTD. Moreover, it has been also shown that a pose consensus approach might also result in database enrichment. We present a new methodology named Pose/Ranking Consensus (PRC) that combines both pose and ranking consensus approaches, to overcome the limitations of each stand-alone strategy. This approach has been developed using four docking programs (ICM, rDock, Auto Dock 4, and PLANTS; the first one is commercial, the other three are free). We undertook a thorough analysis for the best way of combining pose and rank strategies, and applied the PRC to a wide range of 34 targets sampling different protein families and binding site properties. Our approach exhibits an improved systematic performance in terms of enrichment factor and hit rate with respect to either pose consensus or consensus ranking alone strategies at a lower computational cost, while always ensuring the recovery of a suitable number of ligands. An analysis using four free docking programs (replacing ICM by Auto Dock Vina) displayed comparable results.

Functional and druggability analysis of the SARS-CoV-2 proteome.

The infectious coronavirus disease (COVID-19) pandemic, caused by the coronavirus SARS-CoV-2, appeared in December 2019 in Wuhan, China, and has spread worldwide. As of today, more than 46 million people have been infected and over 1.2 million fatalities. With the purpose of contributing to the development of effective therapeutics, we performed an in silico determination of binding hot-spots and an assessment of their druggability within the complete SARS-CoV-2 proteome. All structural, non-structural, and accessory proteins have been studied, and whenever experimental structural data of SARS-CoV-2 proteins were not available, homology models were built based on solved SARS-CoV structures. Several potential allosteric or protein-protein interaction druggable sites on different viral targets were identified, knowledge that could be used to expand current drug discovery endeavors beyond the currently explored cysteine proteases and the polymerase complex. It is our hope that this study will support the efforts of the scientific community both in understanding the molecular determinants of this disease and in widening the repertoire of viral targets in the quest for repurposed or novel drugs against COVID-19.

Bioinformatic analysis of RHO family of GTPases identifies RAC1 pharmacological inhibition as a new therapeutic strategy for hepatocellular carcinoma.

OBJECTIVE: The RHO family of GTPases, particularly RAC1, has been linked with hepatocarcinogenesis, suggesting that their inhibition might be a rational therapeutic approach. We aimed to identify and target deregulated RHO family members in human hepatocellular carcinoma (HCC). DESIGN: We studied expression deregulation, clinical prognosis and transcription programmes relevant to HCC using public datasets. The therapeutic potential of RAC1 inhibitors in HCC was study in vitro and in vivo. RNA-Seq analysis and their correlation with the three different HCC datasets were used to characterise the underlying mechanism on RAC1 inhibition. The therapeutic effect of RAC1 inhibition on liver fibrosis was evaluated. RESULTS: Among the RHO family of GTPases we observed that RAC1 is upregulated, correlates with poor patient survival, and is strongly linked with a prooncogenic transcriptional programme. From a panel of novel RAC1 inhibitors studied, 1D-142 was able to induce apoptosis and cell cycle arrest in HCC cells, displaying a stronger effect in highly proliferative cells. Partial rescue of the RAC1-related oncogenic transcriptional programme was obtained on RAC1 inhibition by 1D-142 in HCC. Most importantly, the RAC1 inhibitor 1D-142 strongly reduce tumour growth and intrahepatic metastasis in HCC mice models. Additionally, 1D-142 decreases hepatic stellate cell activation and exerts an anti-fibrotic effect in vivo. CONCLUSIONS: The bioinformatics analysis of the HCC datasets, allows identifying RAC1 as a new therapeutic target for HCC. The targeted inhibition of RAC1 by 1D-142 resulted in a potent antitumoural effect in highly proliferative HCC established in fibrotic livers.

Flexi-pharma: a molecule-ranking strategy for virtual screening using pharmacophores from ligand-free conformational ensembles.

Computer-aided strategies are useful for reducing the costs and increasing the success-rate in drug discovery. Among these strategies, methods based on pharmacophores (an ensemble of electronic and steric features representing the target active site) are efficient to implement over large compound libraries. However, traditional pharmacophore-based methods require knowledge of active compounds or ligand-receptor structures, and only few ones account for target flexibility. Here, we developed a pharmacophore-based virtual screening protocol, Flexi-pharma, that overcomes these limitations. The protocol uses molecular dynamics (MD) simulations to explore receptor flexibility, and performs a pharmacophore-based virtual screening over a set of MD conformations without requiring prior knowledge about known ligands or ligand-receptor structures for building the pharmacophores. The results from the different receptor conformations are combined using a "voting" approach, where a vote is given to each molecule that matches at least one pharmacophore from each MD conformation. Contrarily to other approaches that reduce the pharmacophore ensemble to some representative models and score according to the matching models or molecule conformers, the Flexi-pharma approach takes directly into account the receptor flexibility by scoring in regards to the receptor conformations. We tested the method over twenty systems, finding an enrichment of the dataset for 19 of them. Flexi-pharma is computationally efficient allowing for the screening of thousands of compounds in minutes on a single CPU core. Moreover, the ranking of molecules by vote is a general strategy that can be applied with any pharmacophore-filtering program.

High-Throughput Docking Using Quantum Mechanical Scoring.

Today high-throughput docking is one of the most commonly used computational tools in drug lead discovery. While there has been an impressive methodological improvement in docking accuracy, docking scoring still remains an open challenge. Most docking programs are rooted in classical molecular mechanics. However, to better characterize protein-ligand interactions, the use of a more accurate quantum mechanical (QM) description would be necessary. In this work, we introduce a QM-based docking scoring function for high-throughput docking and evaluate it on 10 protein systems belonging to diverse protein families, and with different binding site characteristics. Outstanding results were obtained, with our QM scoring function displaying much higher enrichment (screening power) than a traditional docking method. It is acknowledged that developments in quantum mechanics theory, algorithms and computer hardware throughout the upcoming years will allow semi-empirical (or low-cost) quantum mechanical methods to slowly replace force-field calculations. It is thus urgently needed to develop and validate novel quantum mechanical-based scoring functions for high-throughput docking toward more accurate methods for the identification and optimization of modulators of pharmaceutically relevant targets.

Binding Free Energy Calculation Using Quantum Mechanics Aimed for Drug Lead Optimization.

The routine use of in silico tools is already established in drug lead design. Besides the use of molecular docking methods to screen large chemical libraries and thus prioritize compounds for purchase or synthesis, more accurate calculations of protein-ligand binding free energy has shown the potential to guide lead optimization, thus saving time and resources. Theoretical developments and advances in computing power have allowed quantum mechanical-based methods applied to calculations on biomacromolecules to be increasingly explored and used, with the purpose of providing a more accurate description of protein-ligand interactions and an enhanced level of accuracy in the calculation of binding affinities. It should be noted that the quantum mechanical formulation includes, in principle, all contributions to the energy, considering terms usually neglected in molecular mechanics force fields, such as electronic polarization, metal coordination, and covalent binding; moreover, quantum mechanical approaches are systematically improvable. By treating all elements and interactions on equal footing, and avoiding the need of system-dependent parameterizations, they provide a greater degree of transferability. In this review, we illustrate the increasing relevance of quantum mechanical methods for binding free energy calculation in the context of structure-based drug lead optimization, showing representative applications of the different approaches available.

Molecular Docking Using Quantum Mechanical-Based Methods.

Computational methods are a powerful and consolidated tool in the early stage of the drug lead discovery process. Among these techniques, high-throughput molecular docking has proved to be extremely useful in identifying novel bioactive compounds within large chemical libraries. In the docking procedure, the predominant binding mode of each small molecule within a target binding site is assessed, and a docking score reflective of the likelihood of binding is assigned to them. These methods also shed light on how a given hit could be modified in order to improve protein-ligand interactions and are thus able to guide lead optimization. The possibility of reducing time and cost compared to experimental approaches made this technology highly appealing. Due to methodological developments and the increase of computational power, the application of quantum mechanical methods to study macromolecular systems has gained substantial attention in the last decade. A quantum mechanical description of the interactions involved in molecular association of biomolecules may lead to better accuracy compared to molecular mechanics, since there are many physical phenomena that cannot be correctly described within a classical framework, such as covalent bond formation, polarization effects, charge transfer, bond rearrangements, halogen bonding, and others, that require electrons to be explicitly accounted for. Considering the fact that quantum mechanics-based approaches in biomolecular simulation constitute an active and important field of research, we highlight in this work the recent developments of quantum mechanical-based molecular docking and high-throughput docking.

De novo design approaches targeting an envelope protein pocket to identify small molecules against dengue virus.

Dengue fever is a mosquito-borne viral disease that has become a major public health concern worldwide. This disease presents with a wide range of clinical manifestations, from a mild cold-like illness to the more serious hemorrhagic dengue fever and dengue shock syndrome. Currently, neither an approved drug nor an effective vaccine for the treatment are available to fight the disease. The envelope protein (E) is a major component of the virion surface. This protein plays a key role during the viral entry process, constituting an attractive target for the development of antiviral drugs. The crystal structure of the E protein reveals the existence of a hydrophobic pocket occupied by the detergent n-octyl-ß-d-glucoside (ß-OG). This pocket lies at the hinge region between domains I and II and is important for the low pH-triggered conformational rearrangement required for the fusion of the virion with the host's cell. Aiming at the design of novel molecules which bind to E and act as virus entry inhibitors, we undertook a de novo design approach by "growing" molecules inside the hydrophobic site (ß-OG). From more than 240000 small-molecules generated, the 2,4 pyrimidine scaffold was selected as the best candidate, from which one synthesized compound displayed micromolar activity. Molecular dynamics-based optimization was performed on this hit, and thirty derivatives were designed in silico, synthesized and evaluated on their capacity to inhibit dengue virus entry into the host cell. Four compounds were found to be potent antiviral compounds in the low-micromolar range. The assessment of drug-like physicochemical and in vitro pharmacokinetic properties revealed that compounds 3e and 3h presented acceptable solubility values and were stable in mouse plasma, simulated gastric fluid, simulated intestinal fluid, and phosphate buffered saline solution.

Exponential consensus ranking improves the outcome in docking and receptor ensemble docking.

Consensus-scoring methods are commonly used with molecular docking in virtual screening campaigns to filter potential ligands for a protein target. Traditional consensus methods combine results from different docking programs by averaging the score or rank of each molecule obtained from individual programs. Unfortunately, these methods fail if one of the docking programs has poor performance, which is likely to occur due to training-set dependencies and scoring-function parameterization. In this work, we introduce a novel consensus method that overcomes these limitations. We combine the results from individual docking programs using a sum of exponential distributions as a function of the molecule rank for each program. We test the method over several benchmark systems using individual and ensembles of target structures from diverse protein families with challenging decoy/ligand datasets. The results demonstrate that the novel method outperforms the best traditional consensus strategies over a wide range of systems. Moreover, because the novel method is based on the rank rather than the score, it is independent of the score units, scales and offsets, which can hinder the combination of results from different structures or programs. Our method is simple and robust, providing a theoretical basis not only for molecular docking but also for any consensus strategy in general.