Zhang, Zizhuo, Wu, Lijun, Gao, Kaiyuan et al. · ArXiv.org · 2025
Molecular docking that predicts the bound structures of small molecules (ligands) to their protein targets, plays a vital role in drug discovery. However, existing docking methods often face limitations: they either overlook crucial structural changes by assuming protein rigidity or suffer from low computational efficiency due to their reliance on generative models for structure sampling. To address these challenges, we propose FABFlex, a fast and accurate regression-based multi-task learning model designed for realistic blind flexible docking scenarios, where proteins exhibit flexibility and binding pocket sites are unknown (blind). Specifically, FABFlex's architecture comprises three specialized modules working in concert: (1) A pocket prediction module that identifies potential binding sites, addressing the challenges inherent in blind docking scenarios. (2) A ligand docking module that predicts the bound (holo) structures of ligands from their unbound (apo) states. (3) A pocket docking module that forecasts the holo structures of protein pockets from their apo conformations. Notably, FABFlex incorporates an iterative update mechanism that serves as a conduit between the ligand and pocket docking modules, enabling continuous structural refinements. This approach effectively integrates the three subtasks of blind flexible docking-pocket identification, ligand conformation prediction, and protein flexibility modeling-into a unified, coherent framework. Extensive experiments on public benchmark datasets demonstrate that FABFlex not only achieves superior effectiveness in predicting accurate binding modes but also exhibits a significant speed advantage (208 $\times$) compared to existing state-of-the-art methods. Our code is released at https://github.com/tmlr-group/FABFlex.
Cited by 0Open access
Christina Schindler, Sjoerd J. de Vries, Martin Zacharias · Structure · 2015
Cited by 135Open access
Yang Jiang (46459), Charlotte M. Deane (50371), Garrett M. Morris (179999) et al. · Figshare · 2024
<p>The top 5 binding poses are presented for each nascent chain protein structure and each candidate drug. The binding score (Autodock vina score) is shown near each binding pose. The scores for the binding poses that are located on target are colored in red. The protein structures are colored from red to blue from N-terminal tail to C-terminal tail. The C-terminal 20 amino acids in the nascent chains that are embedded in the ribosome exit tunnel were removed in the blind docking.</p> <p>(TIF)</p>
Cited by 0Open access
Yang Liu, Yang Cao · Methods in molecular biology · 2023
Cited by 47
Ryan E. Pavlovicz (205761), Brandon J. Henderson (205764), Andrew B. Bonnell (205767) et al. · Figshare · 2015
<p>Docking modes for epibatidine (A – magenta), nicotine (<b>B</b> – orange), and acetylcholine (<b>C</b> – green) to hα4β2 models compared to crystallographic binding modes (blue). Crystallographic structures for AChBP bound to epibatidine, nicotine, and carbamylcholine (PDB IDs: 2BYN, 1UW6, and 1UV6 respectively) were superimposed on nAChR ECD models to determine RMSDs of the dockings.</p>
Cited by 0Open access
YuPeng Huang (16472792), Hong Zhang (25820), Siyuan Jiang (7889114) et al. · Figshare · 1753
Virtual screening, including molecular docking, plays\nan essential\nrole in drug discovery. Many traditional and machine-learning-based\nmethods are available to fulfill the docking task. However, the traditional\ndocking methods are normally extensively time-consuming, and their\nperformance in blind docking remains to be improved. Although the\nruntime of docking based on machine learning is significantly decreased,\ntheir accuracy is still limited. In this study, we take advantage\nof both traditional and machine-learning-based methods and present\na method, deep site and docking pose (DSDP), to improve the performance\nof blind docking. For traditional blind docking, the entire protein\nis covered by a cube, and the initial positions of ligands are randomly\ngenerated in this cube. In contrast, DSDP can predict the binding\nsite of proteins and provide an accurate searching shape and initial\npositions for further conformational sampling. The sampling task of\nDSDP makes use of the score function and a similar but modified searching\nstrategy of AutoDock Vina, accelerated by implementation in GPUs.\nWe systematically compare its performance in redocking, blind docking,\nand virtual screening tasks with state-of-the-art methods, including\nAutoDock Vina, GNINA, QuickVina, SMINA, and DiffDock. In the blind\ndocking task, DSDP reaches a 29.8% top-1 success rate (root-mean-squared\ndeviation < 2 Å) on an unbiased and challenging test dataset\nwith 1.2 s wall-clock computational time per system. Its performances\non the DUD-E dataset and the time-split PDBBind dataset used in EquiBind,\nTANKBind, and DiffDock are also evaluated, presenting a 57.2 and 41.8%\ntop-1 success rate with 0.8 and 1.0 s per system, respectively.
Cited by 0Open access
Yupeng Huang, Hong Zhang, Siyuan Jiang et al. · Journal of Chemical Information and Modeling · 2023
Virtual screening, including molecular docking, plays an essential role in drug discovery. Many traditional and machine-learning-based methods are available to fulfill the docking task. However, the traditional docking methods are normally extensively time-consuming, and their performance in blind docking remains to be improved. Although the runtime of docking based on machine learning is significantly decreased, their accuracy is still limited. In this study, we take advantage of both traditional and machine-learning-based methods and present a method, deep site and docking pose (DSDP), to improve the performance of blind docking. For traditional blind docking, the entire protein is covered by a cube, and the initial positions of ligands are randomly generated in this cube. In contrast, DSDP can predict the binding site of proteins and provide an accurate searching shape and initial positions for further conformational sampling. The sampling task of DSDP makes use of the score function and a similar but modified searching strategy of AutoDock Vina, accelerated by implementation in GPUs. We systematically compare its performance in redocking, blind docking, and virtual screening tasks with state-of-the-art methods, including AutoDock Vina, GNINA, QuickVina, SMINA, and DiffDock. In the blind docking task, DSDP reaches a 29.8% top-1 success rate (root-mean-squared deviation < 2 Å) on an unbiased and challenging test dataset with 1.2 s wall-clock computational time per system. Its performances on the DUD-E dataset and the time-split PDBBind dataset used in EquiBind, TANKBind, and DiffDock are also evaluated, presenting a 57.2 and 41.8% top-1 success rate with 0.8 and 1.0 s per system, respectively.
Cited by 36
Ian Davis, Kaushik Raha, Martha S. Head et al. · Protein Science · 2009
It is difficult to properly validate algorithms that dock a small molecule ligand into its protein receptor using data from the public domain: the predictions are not blind because the correct binding mode is already known, and public test cases may not be representative of compounds of interest such as drug leads. Here, we use private data from a real drug discovery program to carry out a blind evaluation of the RosettaLigand docking methodology and find that its performance is on average comparable with that of the best commercially available current small molecule docking programs. The strength of RosettaLigand is the use of the Rosetta sampling methodology to simultaneously optimize protein sidechain, protein backbone and ligand degrees of freedom; the extensive benchmark test described here identifies shortcomings in other aspects of the protocol and suggests clear routes to improving the method.
Cited by 85Open access
Avraham Ben‐Shimon, Masha Y. Niv · Structure · 2015
Cited by 81Open access
Yupeng Huang, Hong Zhang, Siyuan Jiang et al. · arXiv (Cornell University) · 2023
Virtual screening, including molecular docking, plays an essential role in drug discovery. Many traditional and machine-learning based methods are available to fulfil the docking task. The traditional docking methods are normally extensively time-consuming, and their performance in blind docking remains to be improved. Although the runtime of docking based on machine learning is significantly decreased, their accuracy is still limited. In this study, we take the advantage of both traditional and machine-learning based methods, and present a method Deep Site and Docking Pose (DSDP) to improve the performance of blind docking. For the traditional blind docking, the entire protein is covered by a cube, and the initial positions of ligands are randomly generated in the cube. In contract, DSDP can predict the binding site of proteins and provide an accurate searching space and initial positions for the further conformational sampling. The docking task of DSDP makes use of the score function and a similar but modified searching strategy of AutoDock Vina, accelerated by implementation in GPUs. We systematically compare its performance with the state-of-the-art methods, including Autodock Vina, GNINA, QuickVina, SMINA, and DiffDock. DSDP reaches a 29.8% top-1 success rate (RMSD < 2 Å) on an unbiased and challenging test dataset with 1.2 s wall-clock computational time per system. Its performances on DUD-E dataset and the time-split PDBBind dataset used in EquiBind, TankBind, and DiffDock are also effective, presenting a 57.2% and 41.8% top-1 success rate with 0.8 s and 1.0 s per system, respectively.
Cited by 2Open access
Muhammad Sohaib Roomi, Giulia Culletta, Lisa Longo et al. · Pharmaceuticals · 2025
Blind docking predicts binding interactions between two molecular entities without prior knowledge of the binding site. This approach is essential because it explores the entire surface of the receptor to identify potential interaction sites. Blind docking widely works for both protein-protein and ligand-protein interaction studies. In protein-protein blind docking, the method aims to predict the correct orientation and interface of two proteins forming a complex. Protein blind docking is particularly valuable in studying transient interactions, protein-protein recognition, signaling pathways, tentative and significant biomolecular assemblies where structural data is limited. Ligand-protein blind docking discovers potential binding pockets across the entire protein surface. It is frequently applied in early-stage drug discovery, especially for novel or poorly characterized targets. The method helps identify allosteric sites or novel binding regions that are not evident from known structures. Overall, blind docking provides a versatile and powerful tool for studying molecular interactions, enabling discovery even in the absence of detailed structural information. In this scenario, we reported a timeline of attempts to improve this kind of computational approach with ML and hybrid approaches to obtain more reliable predictions. We dedicate two main sections to protein-protein and protein-ligand blind docking, presenting the reliability and caveats for each approach and outlining potential future directions.
Cited by 3Open access
Yang Jiang (46459), Charlotte M. Deane (50371), Garrett M. Morris (179999) et al. · Figshare · 2024
<p>The top 5 binding poses are presented for each protein structure and each candidate drug. The binding score (Autodock vina score) is shown near each binding pose. The scores for the binding poses that are located on target are colored in red. The protein structures are colored from red to blue from N-terminal tail to C-terminal tail. Protein structures #3 and #9 have no on-target binding pose detected.</p> <p>(TIF)</p>
Cited by 0Open access
Heike Singer (232666), Arijit Biswas (752609), Nicole Nuesgen (762689) et al. · Figshare · 2015
<p>The figure is split into three panels, each representing the ZBTB16/NLRP7 dock in differently colored formats. The models are depicted in stick form. <b>A.</b> The first panel shows the proteins ZBTB16 and NLRP7 in yellow and green. The interface regions are marked and specified. The red residues on the interface regions belong to ZBTB16, while the blue ones belong to NLRP7. <b>B.</b> and <b>C</b>. Both panels show the same docks but with the individual domains of ZBTB16 and NLRP7, respectively, colored according to the code described below. The apposing protein in each panel is uniformly colored in grey.</p>
Cited by 0Open access
Mónika Bálint, István Horváth, Nikolett Mészáros et al. · International Journal of Molecular Sciences · 2019
Histones serve as protein spools for winding the DNA in the nucleosome. High variability of their post-translational modifications result in a unique code system often responsible for the pathomechanisms of epigenetics-based diseases. Decoding is performed by reader proteins via complex formation with the N-terminal peptide tails of histones. Determination of structures of histone-reader complexes would be a key to unravel the histone code and the design of new drugs. However, the large number of possible histone complex variations imposes a true challenge for experimental structure determination techniques. Calculation of such complexes is difficult due to considerable size and flexibility of peptides and the shallow binding surfaces of the readers. Moreover, location of the binding sites is often unknown, which requires a blind docking search over the entire surface of the target protein. To accelerate the work in this field, a new approach is presented for prediction of the structure of histone H3 peptide tails docked to their targets. Using a fragmenting protocol and a systematic blind docking method, a collection of well-positioned fragments of the H3 peptide is produced. After linking the fragments, reconstitution of anchoring regions of the target-bound H3 peptide conformations was possible. As a first attempt of combination of blind and fragment docking approaches, our new method is named fragment blind docking (FBD).
Cited by 12Open access
Michal Slutzki, Avraham Ben‐Shimon, Masha Y. Niv · Methods in molecular biology · 2017
Cited by 3
Sekkin, Ecem · OpenMETU (Middle East Technical University) · 2024
This thesis explores two distinct biological structures, each presenting unique challenges that cannot be addressed through conventional methods. The first study focuses on the spike protein of SARS-CoV-2 and a peptide-based drug candidate designed by our collaborator, investigating the potential interactions between the protein and the ligand across four different binding sites. The second study examines recently discovered quadruplex-duplex hybrids (PDB ID: 7CV3-7CV4) in the PIM1 oncogene, proposed as potential targets. Known G-quadruplex stabilizer ligands such as Doxorubicin, Braco-19, and RHPS4 are analyzed to provide insights into their binding modes with these quadruplex-duplex hybrids. Molecular docking techniques are enhanced by integrating molecular mechanics calculations using the newly developed software bioAIM from Meddenovo. This software analyzes interaction energies and conformational changes to improve our understanding of ligand-target interactions. Molecular docking studies reveal promising interactions at multiple binding sites on the spike protein, with one site showing significant potential for drug binding. G-quadruplex ligands primarily bind at the Quadruplex-Duplex junction, suggesting the need for further investigation into the impact of G-quadruplex DNA on drug binding. These findings will guide the design and development of new-generation drugs or the re-purposing of existing drugs for these targets.
Cited by 0Open access
Anna Vangone (606191), Safwat Abdel-Azeim (153436), Ivana Caputo (305607) et al. · Figshare · 2015
<p>(<b>a</b>) and (<b>c</b>): the CONS-COCOMAPS consensus maps calculated for the 22 models from the CDR-directed docking simulations (<b>a</b>) and for the 30 models from the ‘blind’ ones (<b>c</b>). Labels have been added for the MB2.8 CDR loops L1-L3 and H1-H3. (<b>b</b>) and (<b>d</b>): Pymol <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0102839#pone.0102839-Delano1" target="_blank">[70]</a> visualization of the representative models of the top ten clusters in both CDR-directed (<b>c</b>) and ‘blind’ docking simulations (<b>d</b>), after superimposition of Ab1-MB2.8. The MB2.8's light chain is colored in pink, its heavy chain is colored in hotpink. All the models using the MB2.8 CDR region for the binding have AIM2 colored in blue, while AIM2 is colored in silver in the 7<sup>th</sup> solution of the blind simulation.</p>
Cited by 0Open access
Joël Janin · Molecular BioSystems · 2010
Abstract Docking algorithms build multimolecular assemblies based on the subunit structures. “Unbound” docking, which starts with the free molecules and allows for conformation changes, may be used to predict the structure of a protein–protein complex. This requires at least two steps, a rigid-body search that determines the relative position and orientation of the subunits, and a refinement step. The methods developed in the past twenty years yield native-like models in most cases, but always with many false positives that must be filtered out, and they fail when the conformation changes are large. CAPRI (Critical Assessment of PRedicted Interactions) is a community-wide experiment set up to monitor progress in the field. It offers participants the opportunity to test their methods in blind predictions that are assessed against an unpublished experimental structure. The models submitted by predictor groups are judged depending on how well they reproduce the geometry and the residue–residue contacts seen in the target structure. In nine years of CAPRI, 42 target complexes have been subjected to prediction based on the components' unbound structures. Good models have been submitted for 28 targets, and prediction has failed on 6. Both these successes and these failures have been fruitful, as they stimulated participant groups to develop new score functions to identify native-like solutions, and new algorithms that allow the molecules to be flexible during docking.
Cited by 165
Marc F. Lensink, Shoshana J. Wodak · Proteins Structure Function and Bioinformatics · 2010
Reliable prediction of the amino acid residues involved in protein-protein interfaces can provide valuable insight into protein function, and inform mutagenesis studies, and drug design applications. A fast-growing number of methods are being proposed for predicting protein interfaces, using structural information, energetic criteria, or sequence conservation or by integrating multiple criteria and approaches. Overall however, their performance remains limited, especially when applied to nonobligate protein complexes, where the individual components are also stable on their own. Here, we evaluate interface predictions derived from protein-protein docking calculations. To this end we measure the overlap between the interfaces in models of protein complexes submitted by 76 participants in CAPRI (Critical Assessment of Predicted Interactions) and those of 46 observed interfaces in 20 CAPRI targets corresponding to nonobligate complexes. Our evaluation considers multiple models for each target interface, submitted by different participants, using a variety of docking methods. Although this results in a substantial variability in the prediction performance across participants and targets, clear trends emerge. Docking methods that perform best in our evaluation predict interfaces with average recall and precision levels of about 60%, for a small majority (60%) of the analyzed interfaces. These levels are significantly higher than those obtained for nonobligate complexes by most extant interface prediction methods. We find furthermore that a sizable fraction (24%) of the interfaces in models ranked as incorrect in the CAPRI assessment are actually correctly predicted (recall and precision ≥50%), and that these models contribute to 70% of the correct docking-based interface predictions overall. Our analysis proves that docking methods are much more successful in identifying interfaces than in predicting complexes, and suggests that these methods have an excellent potential of addressing the interface prediction challenge.
Cited by 81
Gianvito Grasso, Arianna Di Gregorio, Bojan Mavkov et al. · Journal of Biomolecular Structure and Dynamics · 2021
In the present paper we propose a novel blind docking protocol based on Autodock-Vina. The developed docking protocol can provide binding site identification and binding pose prediction at the same time, by a systematical exploration of the protein volume performed with several preliminary docking calculations. In our opinion, this protocol can be successfully applied during the first steps of the virtual screening pipeline, because it provides binding site identification and binding pose prediction at the same time without visual evaluation of the binding site. After the binding pose prediction, MM/GBSA re-scoring rescoring procedures has been applied to improve the accuracy of the protein-ligand bound state. The FRAD protocol has been tested on 116 protein-ligand complexes of the Heat Shock Protein 90 - alpha, on 176 of Human Immunodeficiency virus protease 1, and on more than 100 protein-ligand system taken from the PDBbind dataset. Overall, the FRAD approach combined to MM/GBSA re-scoring can be considered as a powerful tool to increase the accuracy and efficiency with respect to other standard docking approaches when the ligand-binding site is unknown.Communicated by Ramaswamy H. Sarma.
Cited by 28Open access