Binding site matching in rational drug design: algorithms and applications.

Interactions between proteins and small molecules are critical for biological functions. These interactions often occur in small cavities within protein structures, known as ligand-binding pockets. Understanding the physicochemical qualities of binding pockets is essential to improve not only our basic knowledge of biological systems, but also drug development procedures. In order to quantify similarities among pockets in terms of their geometries and chemical properties, either bound ligands can be compared to one another or binding sites can be matched directly. Both perspectives routinely take advantage of computational methods including various techniques to represent and compare small molecules as well as local protein structures. In this review, we survey 12 tools widely used to match pockets. These methods are divided into five categories based on the algorithm implemented to construct binding-site alignments. In addition to the comprehensive analysis of their algorithms, test sets and the performance of each method are described. We also discuss general pharmacological applications of computational pocket matching in drug repurposing, polypharmacology and side effects. Reflecting on the importance of these techniques in drug discovery, in the end, we elaborate on the development of more accurate meta-predictors, the incorporation of protein flexibility and the integration of powerful artificial intelligence technologies such as deep learning.

Estimating the magnitude of near-membrane PDE4 activity in living cells.

Recent studies have demonstrated that functionally discrete pools of phosphodiesterase (PDE) activity regulate distinct cellular functions. While the importance of localized pools of enzyme activity has become apparent, few studies have estimated enzyme activity within discrete subcellular compartments. Here we present an approach to estimate near-membrane PDE activity. First, total PDE activity is measured using traditional PDE activity assays. Second, known cAMP concentrations are dialyzed into single cells and the spatial spread of cAMP is monitored using cyclic nucleotide-gated channels. Third, mathematical models are used to estimate the spatial distribution of PDE activity within cells. Using this three-tiered approach, we observed two pharmacologically distinct pools of PDE activity, a rolipram-sensitive pool and an 8-methoxymethyl IBMX (8MM-IBMX)-sensitive pool. We observed that the rolipram-sensitive PDE (PDE4) was primarily responsible for cAMP hydrolysis near the plasma membrane. Finally, we observed that PDE4 was capable of blunting cAMP levels near the plasma membrane even when 100 µM cAMP were introduced into the cell via a patch pipette. Two compartment models predict that PDE activity near the plasma membrane, near cyclic nucleotide-gated channels, was significantly lower than total cellular PDE activity and that a slow spatial spread of cAMP allowed PDE activity to effectively hydrolyze near-membrane cAMP. These results imply that cAMP levels near the plasma membrane are distinct from those in other subcellular compartments; PDE activity is not uniform within cells; and localized pools of AC and PDE activities are responsible for controlling cAMP levels within distinct subcellular compartments.

GeauxDock: A novel approach for mixed-resolution ligand docking using a descriptor-based force field.

Molecular docking is an important component of computer-aided drug discovery. In this communication, we describe GeauxDock, a new docking approach that builds on the ideas of ligand homology modeling. GeauxDock features a descriptor-based scoring function integrating evolutionary constraints with physics-based energy terms, a mixed-resolution molecular representation of protein-ligand complexes, and an efficient Monte Carlo sampling protocol. To drive docking simulations toward experimental conformations, the scoring function was carefully optimized to produce a correlation between the total pseudoenergy and the native-likeness of binding poses. Indeed, benchmarking calculations demonstrate that GeauxDock has a strong capacity to identify near-native conformations across docking trajectories with the area under receiver operating characteristics of 0.85. By excluding closely related templates, we show that GeauxDock maintains its accuracy at lower levels of homology through the increased contribution from physics-based energy terms compensating for weak evolutionary constraints. GeauxDock is available at http://www.institute.loni.org/lasigma/package/dock/.

eFindSite: improved prediction of ligand binding sites in protein models using meta-threading, machine learning and auxiliary ligands.

Molecular structures and functions of the majority of proteins across different species are yet to be identified. Much needed functional annotation of these gene products often benefits from the knowledge of protein-ligand interactions. Towards this goal, we developed eFindSite, an improved version of FINDSITE, designed to more efficiently identify ligand binding sites and residues using only weakly homologous templates. It employs a collection of effective algorithms, including highly sensitive meta-threading approaches, improved clustering techniques, advanced machine learning methods and reliable confidence estimation systems. Depending on the quality of target protein structures, eFindSite outperforms geometric pocket detection algorithms by 15-40 % in binding site detection and by 5-35 % in binding residue prediction. Moreover, compared to FINDSITE, it identifies 14 % more binding residues in the most difficult cases. When multiple putative binding pockets are identified, the ranking accuracy is 75-78 %, which can be further improved by 3-4 % by including auxiliary information on binding ligands extracted from biomedical literature. As a first across-genome application, we describe structure modeling and binding site prediction for the entire proteome of Escherichia coli. Carefully calibrated confidence estimates strongly indicate that highly reliable ligand binding predictions are made for the majority of gene products, thus eFindSite holds a significant promise for large-scale genome annotation and drug development projects. eFindSite is freely available to the academic community at http://www.brylinski.org/efindsite .

Assessment of cellular mechanisms contributing to cAMP compartmentalization in pulmonary microvascular endothelial cells.

Cyclic AMP signals encode information required to differentially regulate a wide variety of cellular responses; yet it is not well understood how information is encrypted within these signals. An emerging concept is that compartmentalization underlies specificity within the cAMP signaling pathway. This concept is based on a series of observations indicating that cAMP levels are distinct in different regions of the cell. One such observation is that cAMP production at the plasma membrane increases pulmonary microvascular endothelial barrier integrity, whereas cAMP production in the cytosol disrupts barrier integrity. To better understand how cAMP signals might be compartmentalized, we have developed mathematical models in which cellular geometry as well as total adenylyl cyclase and phosphodiesterase activities were constrained to approximate values measured in pulmonary microvascular endothelial cells. These simulations suggest that the subcellular localizations of adenylyl cyclase and phosphodiesterase activities are by themselves insufficient to generate physiologically relevant cAMP gradients. Thus, the assembly of adenylyl cyclase, phosphodiesterase, and protein kinase A onto protein scaffolds is by itself unlikely to ensure signal specificity. Rather, our simulations suggest that reductions in the effective cAMP diffusion coefficient may facilitate the formation of substantial cAMP gradients. We conclude that reductions in the effective rate of cAMP diffusion due to buffers, structural impediments, and local changes in viscosity greatly facilitate the ability of signaling complexes to impart specificity within the cAMP signaling pathway.

Structure-Based Drug Discovery Accelerated by Many-Core Devices.

Computer-aided design is one of the critical components of modern drug discovery. Drug development is routinely streamlined using computational approaches to improve hit identification and lead selection, enhance bioavailability, and reduce toxicity. A mounting body of genomic knowledge accumulated during the last decade or so presents great opportunities for pharmaceutical research. However, new challenges also arose because processing this large volume of data demands unprecedented computing resources. On the other hand, the state-of-the-art heterogeneous systems deliver petaflops of peak performance to accelerate scientific discovery. In this communication, we review modern parallel accelerator architectures, mainly focusing on Intel Xeon Phi many-core devices. Xeon Phi is a relatively new platform that features tens of computing cores with hundreds of threads offering massively parallel capabilities for a broad range of application. We also discuss common parallel programming frameworks targeted to this accelerator, including OpenMP, OpenCL, MPI and HPX. Recent advances in code development for many-core devices are described to demonstrate the advantages of heterogeneous implementations over the traditional, serial computing. Finally, we highlight selected algorithms, eFindSite, a ligand binding site predictor, a force field for bio-molecular simulations, and BUDE, a structure-based virtual screening engine, to demonstrate how modern drug discovery is accelerated by heterogeneous systems equipped with parallel computing devices.

GeauxDock: Accelerating Structure-Based Virtual Screening with Heterogeneous Computing.

Computational modeling of drug binding to proteins is an integral component of direct drug design. Particularly, structure-based virtual screening is often used to perform large-scale modeling of putative associations between small organic molecules and their pharmacologically relevant protein targets. Because of a large number of drug candidates to be evaluated, an accurate and fast docking engine is a critical element of virtual screening. Consequently, highly optimized docking codes are of paramount importance for the effectiveness of virtual screening methods. In this communication, we describe the implementation, tuning and performance characteristics of GeauxDock, a recently developed molecular docking program. GeauxDock is built upon the Monte Carlo algorithm and features a novel scoring function combining physics-based energy terms with statistical and knowledge-based potentials. Developed specifically for heterogeneous computing platforms, the current version of GeauxDock can be deployed on modern, multi-core Central Processing Units (CPUs) as well as massively parallel accelerators, Intel Xeon Phi and NVIDIA Graphics Processing Unit (GPU). First, we carried out a thorough performance tuning of the high-level framework and the docking kernel to produce a fast serial code, which was then ported to shared-memory multi-core CPUs yielding a near-ideal scaling. Further, using Xeon Phi gives 1.9× performance improvement over a dual 10-core Xeon CPU, whereas the best GPU accelerator, GeForce GTX 980, achieves a speedup as high as 3.5×. On that account, GeauxDock can take advantage of modern heterogeneous architectures to considerably accelerate structure-based virtual screening applications. GeauxDock is open-sourced and publicly available at www.brylinski.org/geauxdock and https://figshare.com/articles/geauxdock_tar_gz/3205249.

Calculating an optimal box size for ligand docking and virtual screening against experimental and predicted binding pockets.

BACKGROUND: Computational approaches have emerged as an instrumental methodology in modern research. For example, virtual screening by molecular docking is routinely used in computer-aided drug discovery. One of the critical parameters for ligand docking is the size of a search space used to identify low-energy binding poses of drug candidates. Currently available docking packages often come with a default protocol for calculating the box size, however, many of these procedures have not been systematically evaluated. METHODS: In this study, we investigate how the docking accuracy of AutoDock Vina is affected by the selection of a search space. We propose a new procedure for calculating the optimal docking box size that maximizes the accuracy of binding pose prediction against a non-redundant and representative dataset of 3,659 protein-ligand complexes selected from the Protein Data Bank. Subsequently, we use the Directory of Useful Decoys, Enhanced to demonstrate that the optimized docking box size also yields an improved ranking in virtual screening. Binding pockets in both datasets are derived from the experimental complex structures and, additionally, predicted by eFindSite. RESULTS: A systematic analysis of ligand binding poses generated by AutoDock Vina shows that the highest accuracy is achieved when the dimensions of the search space are 2.9 times larger than the radius of gyration of a docking compound. Subsequent virtual screening benchmarks demonstrate that this optimized docking box size also improves compound ranking. For instance, using predicted ligand binding sites, the average enrichment factor calculated for the top 1 % (10 %) of the screening library is 8.20 (3.28) for the optimized protocol, compared to 7.67 (3.19) for the default procedure. Depending on the evaluation metric, the optimal docking box size gives better ranking in virtual screening for about two-thirds of target proteins. CONCLUSIONS: This fully automated procedure can be used to optimize docking protocols in order to improve the ranking accuracy in production virtual screening simulations. Importantly, the optimized search space systematically yields better results than the default method not only for experimental pockets, but also for those predicted from protein structures. A script for calculating the optimal docking box size is freely available at www.brylinski.org/content/docking-box-size. Graphical AbstractWe developed a procedure to optimize the box size in molecular docking calculations. Left panel shows the predicted binding pose of NADP (green sticks) compared to the experimental complex structure of human aldose reductase (blue sticks) using a default protocol. Right panel shows the docking accuracy using an optimized box size.

Accelerating the Pace of Protein Functional Annotation With Intel Xeon Phi Coprocessors.

Intel Xeon Phi is a new addition to the family of powerful parallel accelerators. The range of its potential applications in computationally driven research is broad; however, at present, the repository of scientific codes is still relatively limited. In this study, we describe the development and benchmarking of a parallel version of eFindSite, a structural bioinformatics algorithm for the prediction of ligand-binding sites in proteins. Implemented for the Intel Xeon Phi platform, the parallelization of the structure alignment portion of eFindSite using pragma-based OpenMP brings about the desired performance improvements, which scale well with the number of computing cores. Compared to a serial version, the parallel code runs 11.8 and 10.1 times faster on the CPU and the coprocessor, respectively; when both resources are utilized simultaneously, the speedup is 17.6. For example, ligand-binding predictions for 501 benchmarking proteins are completed in 2.1 hours on a single Stampede node equipped with the Intel Xeon Phi card compared to 3.1 hours without the accelerator and 36.8 hours required by a serial version. In addition to the satisfactory parallel performance, porting existing scientific codes to the Intel Xeon Phi architecture is relatively straightforward with a short development time due to the support of common parallel programming models by the coprocessor. The parallel version of eFindSite is freely available to the academic community at www.brylinski.org/efindsite.

eFindSite: Enhanced Fingerprint-Based Virtual Screening Against Predicted Ligand Binding Sites in Protein Models.

A standard practice for lead identification in drug discovery is ligand virtual screening, which utilizes computing technologies to detect small compounds that likely bind to target proteins prior to experimental screens. A high accuracy is often achieved when the target protein has a resolved crystal structure; however, using protein models still renders significant challenges. Towards this goal, we recently developed eFindSite that predicts ligand binding sites using a collection of effective algorithms, including meta-threading, machine learning and reliable confidence estimation systems. Here, we incorporate fingerprint-based virtual screening capabilities in eFindSite in addition to its flagship role as a ligand binding pocket predictor. Virtual screening benchmarks using the enhanced Directory of Useful Decoys demonstrate that eFindSite significantly outperforms AutoDock Vina as assessed by several evaluation metrics. Importantly, this holds true regardless of the quality of target protein structures. As a first genome-wide application of eFindSite, we conduct large-scale virtual screening of the entire proteome of Escherichia coli with encouraging results. In the new approach to fingerprint-based virtual screening using remote protein homology, eFindSite demonstrates its compelling proficiency offering a high ranking accuracy and low susceptibility to target structure deformations. The enhanced version of eFindSite is freely available to the academic community at http://www.brylinski.org/efindsite.

Developing eThread pipeline using SAGA-pilot abstraction for large-scale structural bioinformatics.

While most of computational annotation approaches are sequence-based, threading methods are becoming increasingly attractive because of predicted structural information that could uncover the underlying function. However, threading tools are generally compute-intensive and the number of protein sequences from even small genomes such as prokaryotes is large typically containing many thousands, prohibiting their application as a genome-wide structural systems biology tool. To leverage its utility, we have developed a pipeline for eThread--a meta-threading protein structure modeling tool, that can use computational resources efficiently and effectively. We employ a pilot-based approach that supports seamless data and task-level parallelism and manages large variation in workload and computational requirements. Our scalable pipeline is deployed on Amazon EC2 and can efficiently select resources based upon task requirements. We present runtime analysis to characterize computational complexity of eThread and EC2 infrastructure. Based on results, we suggest a pathway to an optimized solution with respect to metrics such as time-to-solution or cost-to-solution. Our eThread pipeline can scale to support a large number of sequences and is expected to be a viable solution for genome-scale structural bioinformatics and structure-based annotation, particularly, amenable for small genomes such as prokaryotes. The developed pipeline is easily extensible to other types of distributed cyberinfrastructure.