Can We Still Trust Docking Results? An Extension of the Applicability of DockBench on PDBbind Database

PoseBusters: AI-based docking methods fail to generate physically valid poses or generalise to novel sequences

The last few years have seen the development of numerous deep learning-based protein-ligand docking methods. They offer huge promise in terms of speed and accuracy. However, despite claims of state-of-the-art performance in terms of crystallographic root-mean-square deviation (RMSD), upon closer inspection, it has become apparent that they often produce physically implausible molecular structures. It is therefore not sufficient to evaluate these methods solely by RMSD to a native binding mode. It is vital, particularly for deep learning-based methods, that they are also evaluated on steric and energetic criteria. We present PoseBusters, a Python package that performs a series of standard quality checks using the well-established cheminformatics toolkit RDKit. The PoseBusters test suite validates chemical and geometric consistency of a ligand including its stereochemistry, and the physical plausibility of intra- and intermolecular measurements such as the planarity of aromatic rings, standard bond lengths, and protein-ligand clashes. Only methods that both pass these checks and predict native-like binding modes should be classed as having "state-of-the-art" performance. We use PoseBusters to compare five deep learning-based docking methods (DeepDock, DiffDock, EquiBind, TankBind, and Uni-Mol) and two well-established standard docking methods (AutoDock Vina and CCDC Gold) with and without an additional post-prediction energy minimisation step using a molecular mechanics force field. We show that both in terms of physical plausibility and the ability to generalise to examples that are distinct from the training data, no deep learning-based method yet outperforms classical docking tools. In addition, we find that molecular mechanics force fields contain docking-relevant physics missing from deep-learning methods. PoseBusters allows practitioners to assess docking and molecular generation methods and may inspire new inductive biases still required to improve deep learning-based methods, which will help drive the development of more accurate and more realistic predictions.

Small-Molecule Inhibitor of Flaviviral NS3-NS5 Interaction with Broad-Spectrum Activity and Efficacy In Vivo

Every year, dengue virus (DENV) causes one hundred million infections worldwide that can result in dengue disease and severe dengue. Two other mosquito-borne flaviviruses, i.e., Zika virus (ZIKV) and West Nile virus (WNV), are responsible of prolonged outbreaks and are associated with severe neurological diseases, congenital defects, and eventually death. These three viruses, despite their importance for global public health, still lack specific drug treatments. Here, we describe the structure-guided discovery of small molecules with pan-flavivirus antiviral potential by a virtual screening of ~1 million structures targeting the NS3-NS5 interaction surface of different flaviviruses. Two molecules inhibited the interaction between DENV NS3 and NS5 in vitro and the replication of all DENV serotypes as well as ZIKV and WNV and exhibited low propensity to select resistant viruses. Remarkably, one molecule demonstrated efficacy in a mouse model of dengue by reducing peak viremia, viral load in target organs, and associated tissue pathology. This study provides the proof of concept that targeting the flaviviral NS3-NS5 interaction is an effective therapeutic strategy able to reduce virus replication in vivo and discloses new chemical scaffolds that could be further developed, thus providing a significant milestone in the development of much awaited broad-spectrum antiflaviviral drugs. IMPORTANCE More than one-third of the human population is at risk of infection by different mosquito-borne flaviviruses. Despite this, no specific antiviral drug is currently available. In this work, using a computational approach based on molecular dynamics simulation and virtual screening of ~1 million small-molecule structures, we identified a compound that targets the interaction between the two sole flaviviral enzymes, i.e., NS3 and NS5. This compound demonstrated pan-serotype anti-DENV activity and pan-flavivirus potential in infected cells, low propensity to select viral resistant mutant viruses, and efficacy in a mouse model of dengue. Broad-spectrum antivirals are much awaited, and this work represents a significant advance toward the development of therapeutic molecules with extended antiflavivirus potential that act by an innovative mechanism and could be used alone or in combination with other antivirals.

Computational strategies to identify new drug candidates against neuroinflammation

The even more increasing application of computational approaches in these last decades has deeply modified the process of discovery and commercialization of new therapeutic entities. This is especially true in the field of neuroinflammation, in which both the peculiar anatomical localization and the presence of the blood-brain barrier makeit mandatory to finely tune the candidates' physicochemical properties from the early stages of the discovery pipeline. The aim of this review is therefore to provide a general overview to the readers about the topic of neuroinflammation, together with the most common computational strategies that can be exploited to discover and design small molecules controlling neuroinflammation, especially those based on the knowledge of the three-dimensional structure of the biological targets of therapeutic interest. The techniques used to describe the molecular recognition mechanisms, such as molecular docking and molecular dynamics, will therefore be eviscerated, highlighting their advantages and their limitations. Finally, we report several case studies in which computational methods have been applied in drug discovery on neuroinflammation, focusing on the last decade's research.

Re-Exploring the Ability of Common Docking Programs to Correctly Reproduce the Binding Modes of Non-Covalent Inhibitors of SARS-CoV-2 Protease Mpro

In the latest few decades, molecular docking has imposed itself as one of the most used approaches for computational drug discovery. Several docking benchmarks have been published, comparing the performance of different algorithms in respect to a molecular target of interest, usually evaluating their ability in reproducing the experimental data, which, in most cases, comes from X-ray structures. In this study, we elucidated the variation of the performance of three docking algorithms, namely GOLD, Glide, and PLANTS, in replicating the coordinates of the crystallographic ligands of SARS-CoV-2 main protease (Mpro). Through the comparison of the data coming from docking experiments and the values derived from the calculation of the solvent exposure of the crystallographic ligands, we highlighted the importance of this last variable for docking performance. Indeed, we underlined how an increase in the percentage of the ligand surface exposed to the solvent in a crystallographic complex makes it harder for the docking algorithms to reproduce its conformation. We further validated our hypothesis through molecular dynamics simulations, showing that the less stable protein–ligand complexes (in terms of root-mean-square deviation and root-mean-square fluctuation) tend to be derived from the cases in which the solvent exposure of the ligand in the starting system is higher.

Fine tuning for success in structure-based virtual screening

Structure-based virtual screening plays a significant role in drug-discovery. The method virtually docks millions of compounds from corporate or public libraries into a binding site of a disease-related protein structure, allowing for the selection of a small list of potential ligands for experimental testing. Many algorithms are available for docking and assessing the affinity of compounds for a targeted protein site. The performance of affinity estimation calculations is highly dependent on the size and nature of the site, therefore a rationale for selecting the best protocol is required. To address this issue, we have developed an automated calibration process, implemented in a Knime workflow. It consists of four steps: preparation of a protein test set with structures and models of the target, preparation of a compound test set with target-related ligands and decoys, automatic test of 24 scoring/rescoring protocols for each target structure and model, and graphical display of results. The automation of the process combined with execution on high performance computing resources greatly reduces the duration of the calibration phase, and the test of many combinations of algorithms on various target conformations results in a rational and optimal choice of the best protocol. Here, we present this tool and exemplify its application in setting-up an optimal protocol for SBVS against Retinoid X Receptor alpha.

Comparative study of the interaction of ivermectin with proteins of interest associated with SARS-CoV-2: A computational and biophysical approach

The SARS-CoV-2 pandemic has accelerated the study of existing drugs. The mixture of homologs called ivermectin (avermectin-B1a [HB1a] + avermectin-B1b [HB1b]) has shown antiviral activity against SARS-CoV-2 in vitro. However, there are few reports on the behavior of each homolog. We investigated the interaction of each homolog with promising targets of interest associated with SARS-CoV-2 infection from a biophysical and computational-chemistry perspective using docking and molecular dynamics. We observed a differential behavior for each homolog, with an affinity of HB1b for viral structures, and of HB1a for host structures considered. The induced disturbances were differential and influenced by the hydrophobicity of each homolog and of the binding pockets. We present the first comparative analysis of the potential theoretical inhibitory effect of both avermectins on biomolecules associated with COVID-19, and suggest that ivermectin through its homologs, has a multiobjective behavior.

Combination of consensus and ensemble docking strategies for the discovery of human dihydroorotate dehydrogenase inhibitors

The inconsistencies in the performance of the virtual screening (VS) process, depending on the used software and structural conformation of the protein, is a challenging issue in the drug design and discovery field. Varying performance, especially in terms of early recognition of the potential hit compounds, negatively affects the whole process and leads to unnecessary waste of the time and resources. Appropriate application of the ensemble docking and consensus-scoring approaches can significantly increase reliability of the VS results. Dihydroorotate dehydrogenase (DHODH) is a key enzyme in the pyrimidine biosynthesis pathway. It is considered as a valuable therapeutic target in cancer, autoimmune and viral diseases. Based on the conducted benchmark study and analysis of the effect of different combinations of the applied methods and approaches, here we suggested a structure-based virtual screening (SBVS) workflow that can be used to increase the reliability of VS.

A Deep-Learning Approach toward Rational Molecular Docking Protocol Selection

While a plethora of different protein–ligand docking protocols have been developed over the past twenty years, their performances greatly depend on the provided input protein–ligand pair. In this study, we developed a machine-learning model that uses a combination of convolutional and fully connected neural networks for the task of predicting the performance of several popular docking protocols given a protein structure and a small compound. We also rigorously evaluated the performance of our model using a widely available database of protein–ligand complexes and different types of data splits. We further open-source all code related to this study so that potential users can make informed selections on which protocol is best suited for their particular protein–ligand pair.