Astex Diverse Set Research Articles

Improved docking of peptides and small molecules in iMOLSDOCK.

iMOLSDOCK is an induced-fit docking algorithm that uses the mutually orthogonal Latin squares (MOLS) sampling technique. Here, we describe the updates made to iMOLSDOCK in order to increase receptor flexibility, improve the scoring system, and speed up calculation. With a dataset of 35 peptide-protein complexes, the PepSet benchmark dataset of 80 peptide-protein complexes, and the Astex Diverse set, which uses nonpeptide small molecules as ligands, iMOLSDOCK has been benchmarked and validated. Flexible residues are now able to deviate from the starting position by a maximum of 3.0Å due to the increased receptor flexibility. The ranking effectiveness of iMOLSDOCK has increased by 24% once the scoring system was improved. Additionally, iMOLSDOCK has been compared to Gold v5.2.1, HPEPDOCK, AutoDock CrankPep v1.0, AutoDock Vina, HADDOCK, PatchDock, and RosettaLigand. For induced-fit peptide-protein docking, iMOLSDOCK achieved success rates of 6%, 37%, and 89% at the top 1, 10, and 100 levels. At the top 1, 10, and 100 levels, iMOLSDOCK had success rates for small molecule-protein docking of 14%, 31%, and 49%. The computation time for peptide docking was lowered by two orders of magnitude, and for nonpeptide small molecule docking, it was roughly 14 times faster due to code optimization in the iMOLSDOCK docking tool. Source code and binary of iMOLSDOCK could be obtained from https://sourceforge.net/projects/mols2-0/files/ .

SMPLIP-Score: predicting ligand binding affinity from simple and interpretable on-the-fly interaction fingerprint pattern descriptors

In drug discovery, rapid and accurate prediction of protein–ligand binding affinities is a pivotal task for lead optimization with acceptable on-target potency as well as pharmacological efficacy. Furthermore, researchers hope for a high correlation between docking score and pose with key interactive residues, although scoring functions as free energy surrogates of protein–ligand complexes have failed to provide collinearity. Recently, various machine learning or deep learning methods have been proposed to overcome the drawbacks of scoring functions. Despite being highly accurate, their featurization process is complex and the meaning of the embedded features cannot directly be interpreted by human recognition without an additional feature analysis. Here, we propose SMPLIP-Score (Substructural Molecular and Protein–Ligand Interaction Pattern Score), a direct interpretable predictor of absolute binding affinity. Our simple featurization embeds the interaction fingerprint pattern on the ligand-binding site environment and molecular fragments of ligands into an input vectorized matrix for learning layers (random forest or deep neural network). Despite their less complex features than other state-of-the-art models, SMPLIP-Score achieved comparable performance, a Pearson’s correlation coefficient up to 0.80, and a root mean square error up to 1.18 in pK units with several benchmark datasets (PDBbind v.2015, Astex Diverse Set, CSAR NRC HiQ, FEP, PDBbind NMR, and CASF-2016). For this model, generality, predictive power, ranking power, and robustness were examined using direct interpretation of feature matrices for specific targets.

Deep Learning in Drug Design: Protein-Ligand Binding Affinity Prediction.

Computational drug design relies on the calculation of binding strength between two biological counterparts especially a chemical compound, i.e., a ligand, and a protein. Predicting the affinity of protein-ligand binding with reasonable accuracy is crucial for drug discovery, and enables the optimization of compounds to achieve better interaction with their target protein. In this paper, we propose a data-driven framework named DeepAtom to accurately predict the protein-ligand binding affinity. With 3D Convolutional Neural Network (3D-CNN) architecture, DeepAtom could automatically extract binding related atomic interaction patterns from the voxelized complex structure. Compared with the other CNN based approaches, our light-weight model design effectively improves the model representational capacity, even with the limited available training data. We carried out validation experiments on the PDBbind v.2016 benchmark and the independent Astex Diverse Set. We demonstrate that the less feature engineering dependent DeepAtom approach consistently outperforms the other baseline scoring methods. We also compile and propose a new benchmark dataset to further improve the model performances. With the new dataset as training input, DeepAtom achieves Pearson's R=0.83 and RMSE=1.23 pK units on the PDBbind v.2016 core set. The promising results demonstrate that DeepAtom models can be potentially adopted in computational drug development protocols such as molecular docking and virtual screening.

High-throughput quantum-mechanics/molecular-mechanics (ONIOM) macromolecular crystallographic refinement with PHENIX/DivCon: the impact of mixed Hamiltonian methods on ligand and protein structure.

Conventional macromolecular crystallographic refinement relies on often dubious stereochemical restraints, the preparation of which often requires human validation for unusual species, and on rudimentary energy functionals that are devoid of nonbonding effects owing to electrostatics, polarization, charge transfer or even hydrogen bonding. While this approach has served the crystallographic community for decades, as structure-based drug design/discovery (SBDD) has grown in prominence it has become clear that these conventional methods are less rigorous than they need to be in order to produce properly predictive protein-ligand models, and that the human intervention that is required to successfully treat ligands and other unusual chemistries found in SBDD often precludes high-throughput, automated refinement. Recently, plugins to the Python-based Hierarchical ENvironment for Integrated Xtallography (PHENIX) crystallographic platform have been developed to augment conventional methods with the in situ use of quantum mechanics (QM) applied to ligand(s) along with the surrounding active site(s) at each step of refinement [Borbulevych et al. (2014), Acta Cryst D70, 1233-1247]. This method (Region-QM) significantly increases the accuracy of the X-ray refinement process, and this approach is now used, coupled with experimental density, to accurately determine protonation states, binding modes, ring-flip states, water positions and so on. In the present work, this approach is expanded to include a more rigorous treatment of the entire structure, including the ligand(s), the associated active site(s) and the entire protein, using a fully automated, mixed quantum-mechanics/molecular-mechanics (QM/MM) Hamiltonian recently implemented in the DivCon package. This approach was validated through the automatic treatment of a population of 80 protein-ligand structures chosen from the Astex Diverse Set. Across the entire population, this method results in an average 3.5-fold reduction in ligand strain and a 4.5-fold improvement in MolProbity clashscore, as well as improvements in Ramachandran and rotamer outlier analyses. Overall, these results demonstrate that the use of a structure-wide QM/MM Hamiltonian exhibits improvements in the local structural chemistry of the ligand similar to Region-QM refinement but with significant improvements in the overall structure beyond the active site.

A Hybrid Cuckoo Search and Differential Evolution Approach to Protein⁻Ligand Docking.

Protein–ligand docking is a molecular modeling technique that is used to predict the conformation of a small molecular ligand at the binding pocket of a protein receptor. There are many protein–ligand docking tools, among which AutoDock Vina is the most popular open-source docking software. In recent years, there have been numerous attempts to optimize the search process in AutoDock Vina by means of heuristic optimization methods, such as genetic and particle swarm optimization algorithms. This study, for the first time, explores the use of cuckoo search (CS) to solve the protein–ligand docking problem. The result of this study is CuckooVina, an enhanced conformational search algorithm that hybridizes cuckoo search with differential evolution (DE). Extensive tests using two benchmark datasets, PDBbind 2012 and Astex Diverse set, show that CuckooVina improves the docking performances in terms of RMSD, binding affinity, and success rate compared to Vina though it requires about 9–15% more time to complete a run than Vina. CuckooVina predicts more accurate docking poses with higher binding affinities than PSOVina with similar success rates. CuckooVina’s slower convergence but higher accuracy suggest that it is better able to escape from local energy minima and improves the problem of premature convergence. As a summary, our results assure that the hybrid CS–DE process to continuously generate diverse solutions is a good strategy to maintain the proper balance between global and local exploitation required for the ligand conformational search.

LigVoxel: inpainting binding pockets using 3D-convolutional neural networks.

Structure-based drug discovery methods exploit protein structural information to design small molecules binding to given protein pockets. This work proposes a purely data driven, structure-based approach for imaging ligands as spatial fields in target protein pockets. We use an end-to-end deep learning framework trained on experimental protein-ligand complexes with the intention of mimicking a chemist's intuition at manually placing atoms when designing a new compound. We show that these models can generate spatial images of ligand chemical properties like occupancy, aromaticity and donor-acceptor matching the protein pocket. The predicted fields considerably overlap with those of unseen ligands bound to the target pocket. Maximization of the overlap between the predicted fields and a given ligand on the Astex diverse set recovers the original ligand crystal poses in 70 out of 85 cases within a threshold of 2 Å RMSD. We expect that these models can be used for guiding structure-based drug discovery approaches. LigVoxel is available as part of the PlayMolecule.org molecular web application suite. Supplementary data are available at Bioinformatics online.

Development and evaluation of a deep learning model for protein-ligand binding affinity prediction.

MotivationStructure based ligand discovery is one of the most successful approaches for augmenting the drug discovery process. Currently, there is a notable shift towards machine learning (ML) methodologies to aid such procedures. Deep learning has recently gained considerable attention as it allows the model to ‘learn’ to extract features that are relevant for the task at hand.ResultsWe have developed a novel deep neural network estimating the binding affinity of ligand–receptor complexes. The complex is represented with a 3D grid, and the model utilizes a 3D convolution to produce a feature map of this representation, treating the atoms of both proteins and ligands in the same manner. Our network was tested on the CASF-2013 ‘scoring power’ benchmark and Astex Diverse Set and outperformed classical scoring functions.Availability and implementationThe model, together with usage instructions and examples, is available as a git repository at http://gitlab.com/cheminfIBB/pafnucy.Supplementary information Supplementary data are available at Bioinformatics online.

GalaxyDock BP2 score: a hybrid scoring function for accurate protein-ligand docking.

Protein-ligand docking is a useful tool for providing atomic-level understanding of protein functions in nature and design principles for artificial ligands or proteins with desired properties. The ability to identify the true binding pose of a ligand to a target protein among numerous possible candidate poses is an essential requirement for successful protein-ligand docking. Many previously developed docking scoring functions were trained to reproduce experimental binding affinities and were also used for scoring binding poses. However, in this study, we developed a new docking scoring function, called GalaxyDock BP2 Score, by directly training the scoring power of binding poses. This function is a hybrid of physics-based, empirical, and knowledge-based score terms that are balanced to strengthen the advantages of each component. The performance of the new scoring function exhibits significant improvement over existing scoring functions in decoy pose discrimination tests. In addition, when the score is used with the GalaxyDock2 protein-ligand docking program, it outperformed other state-of-the-art docking programs in docking tests on the Astex diverse set, the Cross2009 benchmark set, and the Astex non-native set. GalaxyDock BP2 Score and GalaxyDock2 with this score are freely available at http://galaxy.seoklab.org/softwares/galaxydock.html .

On-the-Fly QM/MM Docking with Attracting Cavities.

We developed a hybrid quantum mechanical/molecular mechanical (QM/MM) on-the-fly docking algorithm to address the challenges of treating polarization and selected metal interactions in docking. The algorithm is based on our classical docking algorithm Attracting Cavities and relies on the semiempirical self-consistent charge density functional tight-binding (SCC-DFTB) method and the CHARMM force field. We benchmarked the performance of this approach on three very diverse data sets: (1) the Astex Diverse set of 85 common noncovalent drug/target complexes formed both by hydrophobic and electrostatic interactions; (2) a zinc metalloprotein data set of 281 complexes, where polarization is strong and ligand/protein interactions are dominated by electrostatic interactions; and (3) a heme protein data set of 72 complexes, where ligand/protein interactions are dominated by covalent ligand/iron binding. Redocking performance of the on-the-fly QM/MM docking algorithm was compared to the performance of classical Attracting Cavities, AutoDock, AutoDock Vina, and GOLD. The results demonstrate that the QM/MM code preserves the high accuracy of most classical scores on the Astex Diverse set, while it yields significant improvements on both sets of metalloproteins at moderate computational cost.

AutoSite: an automated approach for pseudo-ligands prediction-from ligand-binding sites identification to predicting key ligand atoms.

The identification of ligand-binding sites from a protein structure facilitates computational drug design and optimization, and protein function assignment. We introduce AutoSite: an efficient software tool for identifying ligand-binding sites and predicting pseudo ligand corresponding to each binding site identified. Binding sites are reported as clusters of 3D points called fills in which every point is labelled as hydrophobic or as hydrogen bond donor or acceptor. From these fills AutoSite derives feature points: a set of putative positions of hydrophobic-, and hydrogen-bond forming ligand atoms. We show that AutoSite identifies ligand-binding sites with higher accuracy than other leading methods, and produces fills that better matches the ligand shape and properties, than the fills obtained with a software program with similar capabilities, AutoLigand In addition, we demonstrate that for the Astex Diverse Set, the feature points identify 79% of hydrophobic ligand atoms, and 81% and 62% of the hydrogen acceptor and donor hydrogen ligand atoms interacting with the receptor, and predict 81.2% of water molecules mediating interactions between ligand and receptor. Finally, we illustrate potential uses of the predicted feature points in the context of lead optimization in drug discovery projects. http://adfr.scripps.edu/AutoDockFR/autosite.html CONTACT: sanner@scripps.eduSupplementary information: Supplementary data are available at Bioinformatics online.

GLIMPSED: An Improved Docking Protocol for Cognate Docking

The cognate docking performance of Glide was evaluated using the Astex diverse set. The standard Glide SP protocol obtained a 85.7 % success rate when the shape similarity cutoff was set to 0.625. The analysis of docking failures pointed out that, when the SiteMap binding site exposure is less than 0.491, 51 out of 52 ligands are correctly positioned. In light of this important finding, an improved docking protocol called GLIMPSED was set up. GLIMPSED relies on the standard Glide SP protocol for binding sites with low exposure, while a more accurate sampling of docking poses followed by a final MM-GBSA rescoring is performed for those highly exposed to the solvent. GLIMPSED and Glide SP were compared applying an accurate and rigorous metrics described in the literature by Hawkins et al. As a result, GLIMPSED was able to dock 89.3 % of the ligands in a correct way. Even though this number is not remarkably different from that obtained with the standard method, it is indeed statistically significant. Applying the Cohen's effect size test, a small, but not trivial, superiority of GLIMPSED was found with respect to standard Glide SP protocol. In our view the SiteMap binding site exposure parameter should be used as guideline to decide whether either a standard or a more extended docking protocol has to be adopted to predict good binding poses.

The Impact of Experimental, Protein Structure on our Ability to Model Protein Function

Gaining an accurate understanding of 3D protein:ligand complex structure, including proper protonation and explicit solvent effects, is crucial for obtaining meaningful results from docking, molecular dynamics, thermodynamic calculations, active site exploration, and ultimately lead optimization in structure based drug discovery (SBDD). In this work, we measure the impact of improved experimental structure - available with the advent of modern, quantum mechanics-based X-ray crystallographic refinement methods - on our understanding of protein function. We will discuss results from the study of docking/scoring (i.e. binding affinity prediction) as a function of structure quality as obtained through the characterization and re-refinement of several, “standard” highly-curated data sets including the Astex Diverse Set, CSAR, Iridium, and others. For each considered set, we have refined the associated structures using advanced, chemically rigorous, QM/MM crystallographic methods, explored key protonation states, and chosen the conformer which most closely matches the experimental density. The calculated binding affinities both before and after re-refinement have been determined using “name brand” score functions, and these calculated affinities have been compared to experimental binding affinity data in order to measure the impact improved structure on in silico characterization. In addition to the impact of structure as it pertains to affinity prediction, improvements in experimental density agreement (e.g. RSR, RCC, and ZDD) and in ligand strain will also be discussed.

AutoDockFR: Advances in Protein-Ligand Docking with Explicitly Specified Binding Site Flexibility.

Automated docking of drug-like molecules into receptors is an essential tool in structure-based drug design. While modeling receptor flexibility is important for correctly predicting ligand binding, it still remains challenging. This work focuses on an approach in which receptor flexibility is modeled by explicitly specifying a set of receptor side-chains a-priori. The challenges of this approach include the: 1) exponential growth of the search space, demanding more efficient search methods; and 2) increased number of false positives, calling for scoring functions tailored for flexible receptor docking. We present AutoDockFR–AutoDock for Flexible Receptors (ADFR), a new docking engine based on the AutoDock4 scoring function, which addresses the aforementioned challenges with a new Genetic Algorithm (GA) and customized scoring function. We validate ADFR using the Astex Diverse Set, demonstrating an increase in efficiency and reliability of its GA over the one implemented in AutoDock4. We demonstrate greatly increased success rates when cross-docking ligands into apo receptors that require side-chain conformational changes for ligand binding. These cross-docking experiments are based on two datasets: 1) SEQ17 –a receptor diversity set containing 17 pairs of apo-holo structures; and 2) CDK2 –a ligand diversity set composed of one CDK2 apo structure and 52 known bound inhibitors. We show that, when cross-docking ligands into the apo conformation of the receptors with up to 14 flexible side-chains, ADFR reports more correctly cross-docked ligands than AutoDock Vina on both datasets with solutions found for 70.6% vs. 35.3% systems on SEQ17, and 76.9% vs. 61.5% on CDK2. ADFR also outperforms AutoDock Vina in number of top ranking solutions on both datasets. Furthermore, we show that correctly docked CDK2 complexes re-create on average 79.8% of all pairwise atomic interactions between the ligand and moving receptor atoms in the holo complexes. Finally, we show that down-weighting the receptor internal energy improves the ranking of correctly docked poses and that runtime for AutoDockFR scales linearly when side-chain flexibility is added.

Attracting cavities for docking. Replacing the rough energy landscape of the protein by a smooth attracting landscape.

Molecular docking is a computational approach for predicting the most probable position of ligands in the binding sites of macromolecules and constitutes the cornerstone of structure‐based computer‐aided drug design. Here, we present a new algorithm called Attracting Cavities that allows molecular docking to be performed by simple energy minimizations only. The approach consists in transiently replacing the rough potential energy hypersurface of the protein by a smooth attracting potential driving the ligands into protein cavities. The actual protein energy landscape is reintroduced in a second step to refine the ligand position. The scoring function of Attracting Cavities is based on the CHARMM force field and the FACTS solvation model. The approach was tested on the 85 experimental ligand–protein structures included in the Astex diverse set and achieved a success rate of 80% in reproducing the experimental binding mode starting from a completely randomized ligand conformer. The algorithm thus compares favorably with current state‐of‐the‐art docking programs. © 2015 The Authors. Journal of Computational Chemistry Published by Wiley Periodicals, Inc.

Facing the Challenges of Structure-Based Target Prediction by Inverse Virtual Screening

Computational target prediction for bioactive compounds is a promising field in assessing off-target effects. Structure-based methods not only predict off-targets, but, simultaneously, binding modes, which are essential for understanding the mode of action and rationally designing selective compounds. Here, we highlight the current open challenges of computational target prediction methods based on protein structures and show why inverse screening rather than sequential pairwise protein-ligand docking methods are needed. A new inverse screening method based on triangle descriptors is introduced: iRAISE (inverse Rapid Index-based Screening Engine). A Scoring Cascade considering the reference ligand as well as the ligand and active site coverage is applied to overcome interprotein scoring noise of common protein-ligand scoring functions. Furthermore, a statistical evaluation of a score cutoff for each individual protein pocket is used. The ranking and binding mode prediction capabilities are evaluated on different datasets and compared to inverse docking and pharmacophore-based methods. On the Astex Diverse Set, iRAISE ranks more than 35% of the targets to the first position and predicts more than 80% of the binding modes with a root-mean-square deviation (RMSD) accuracy of <2.0 Å. With a median computing time of 5 s per protein, large amounts of protein structures can be screened rapidly. On a test set with 7915 protein structures and 117 query ligands, iRAISE predicts the first true positive in a ranked list among the top eight ranks (median), i.e., among 0.28% of the targets.

Protoss: a holistic approach to predict tautomers and protonation states in protein-ligand complexes

The calculation of hydrogen positions is a common preprocessing step when working with crystal structures of protein-ligand complexes. An explicit description of hydrogen atoms is generally needed in order to analyze the binding mode of particular ligands or to calculate the associated binding energies. Due to the large number of degrees of freedom resulting from different chemical moieties and the high degree of mutual dependence this problem is anything but trivial. In addition to an efficient algorithm to take care of the complexity resulting from complicated hydrogen bonding networks, a robust chemical model is needed to describe effects such as tautomerism and ionization consistently. We present a novel method for the placement of hydrogen coordinates in protein-ligand complexes which takes tautomers and protonation states of both protein and ligand into account. Our method generates the most probable hydrogen positions on the basis of an optimal hydrogen bonding network using an empirical scoring function. The high quality of our results could be verified by comparison to the manually adjusted Astex diverse set and a remarkably low rate of undesirable hydrogen contacts compared to other tools.

Self-adaptive differential evolution algorithm incorporating local search for protein-ligand docking

Differential Evolution (DE) algorithm is powerful in optimization problems over several real parameters. DE depends on strategies to generate new trial solutions and the associated parameter values for searching performance. In self-adaptive DE, the automatic learning about previous evolution was used to determine the best mutation strategy and its parameter settings. By combining the self-adaptive DE and Hooke Jeeves local search, we developed a new docking method named SADock (Strategy Adaptation Dock) with the help of AutoDock4 scoring function. As the accuracy and performance of SADock was evaluated in self-docking using the Astex diverse set, the introduced SADock showed better success ratio (89%) than the success ratio (60%) of the Lamarckian genetic algorithm (LGA) of AutoDock4. The self-adapting scheme enabled our new docking method to converge fast and to be robust through the various docking problems.

A consistent description of HYdrogen bond and DEhydration energies in protein–ligand complexes: methods behind the HYDE scoring function

The estimation of free energy of binding is a key problem in structure-based design. We developed the scoring function HYDE based on a consistent description of HYdrogen bond and DEhydration energies in protein-ligand complexes. HYDE is applicable to all types of protein targets since it is not calibrated on experimental binding affinity data or protein-ligand complexes. The comprehensible atom-based score of HYDE is visualized by applying a very intuitive coloring scheme, thereby facilitating the analysis of protein-ligand complexes in the lead optimization process. In this paper, we have revised several aspects of the former version of HYDE which was described in detail previously. The revised HYDE version was already validated in large-scale redocking and screening experiments which were performed in the course of the Docking and Scoring Symposium at 241st ACS National Meeting. In this study, we additionally evaluate the ability of the revised HYDE version to predict binding affinities. On the PDBbind 2007 coreset, HYDE achieves a correlation coefficient of 0.62 between the experimental binding constants and the predicted binding energy, performing second best on this dataset compared to 17 other well-established scoring functions. Further, we show that the performance of HYDE in large-scale redocking and virtual screening experiments on the Astex diverse set and the DUD dataset respectively, is comparable to the best methods in this field.

S4MPLE – Sampler For Multiple Protein–Ligand Entities: Simultaneous Docking of Several Entities

S4MPLE is a conformational sampling tool, based on a hybrid genetic algorithm, simulating one (conformer enumeration) or more molecules (docking). Energy calculations are based on the AMBER force field [Cornell et al. J. Am. Chem. Soc. 1995, 117, 5179.] for biological macromolecules and its generalized version GAFF [Wang et al. J. Comput. Chem. 2004 , 25, 1157.] for ligands. This paper describes more advanced, specific applications of S4MPLE to problems more complex than classical redocking of drug-like compounds [Hoffer et al. J. Mol. Graphics Modell. 2012, submitted for publication.]. Here, simultaneous docking of multiple entities is addressed in two different important contexts. First, simultaneous docking of two fragment-like ligands was attempted, as such ternary complexes are the basis of fragment-based drug design by linkage of the independent binders. As a preliminary, the capacity of S4MPLE to dock fragment-like compounds has been assessed, since this class of small probes used in fragment-based drug design covers a different chemical space than drug-like molecules. Herein reported success rates from fragments redocking are as good as classical benchmarking results on drug-like compounds (Astex Diverse Set [Hartshorn et al. J. Med. Chem. 2007, 50, 726.]). Then, S4MPLE is successfully challenged to predict locations of fragments involved in ternary complexes by means of multientity docking. Second, the key problem of predicting water-mediated interaction is addressed by considering explicit water molecules as additional entities to be docked in the presence of the "main" ligand. Blind prediction of solvent molecule positions, reproducing relevant ligand-water-site mediated interactions, is achieved in 76% cases over saved poses. S4MPLE was also successful to predict crystallographic water displacement by a therefore tailored functional group in the optimized ligand. However, water localization is an extremely delicate issue in terms of weighing of electrostatic and desolvation terms and also introduces a significant increase of required sampling efforts. Yet, the herein reported results - not making use of massively parallel deployment of the software - are very encouraging.

Identification of Ligand Templates using Local Structure Alignment for Structure-Based Drug Design

With a rapid increase in the number of high-resolution protein-ligand structures, the known protein-ligand structures can be used to gain insight into ligand-binding modes in a target protein. On the basis of the fact that the structurally similar binding sites share information about their ligands, we have developed a local structure alignment tool, G-LoSA (graph-based local structure alignment). The known protein-ligand binding-site structure library is searched by G-LoSA to detect binding-site structures with similar geometry and physicochemical properties to a query binding-site structure regardless of sequence continuity and protein fold. Then, the ligands in the identified complexes are used as templates (i.e., template ligands) to predict/design a ligand for the target protein. The performance of G-LoSA is validated against 76 benchmark targets from the Astex diverse set. Using the currently available protein-ligand structure library, G-LoSA is able to identify a single template ligand (from a nonhomologous protein complex) that is highly similar to the target ligand in more than half of the benchmark targets. In addition, our benchmark analyses show that an assembly of structural fragments from multiple template ligands with partial similarity to the target ligand can be used to design novel ligand structures specific to the target protein. This study clearly indicates that a template-based ligand modeling has potential for de novo ligand design and can be a complementary approach to the receptor structure based methods.