Research projects

One of the increasingly important component of modern drug discovery is the prediction of the toxicity of drug candidates. In collaboration with Supratik Mukhopadhyay and Hsiao-Chun Wu, we developed eToxPred, a new approach to reliably estimate the toxicity and synthetic accessibility of small organic compounds. eToxPred employs machine learning algorithms trained on molecular fingerprints to evaluate drug candidates. Encouragingly, it predicts the synthetic accessibility with the mean square error of only 4% and the toxicity with the accuracy of as high as 72%. eToxPred is a valuable tool that can be employed at the outset of drug discovery to filter out those drug candidates that are potentially toxic or would be difficult to synthesize.

Aromatic stacking has long been recognized as one of the key constituents of ligand-protein interfaces. We developed a two-parameter geometric model to analyze aromatic contacts in the experimental and computer-generated structures of ligand-protein complexes, considering various combinations of aromatic amino acid residues and ligand rings. Although modeling aromatic stacking with van der Waals and Coulombic potentials generally provides a sufficient specificity, the geometry of π-π contacts in high-scoring docking conformations could still be improved. The comprehensive analysis of aromatic geometries at ligand-protein interfaces lies the foundation for the development of type-specific statistical potentials to more accurately describe aromatic interactions in molecular docking.

About 7,000 rare, or orphan, diseases affect more than 350 million people worldwide. Although these conditions collectively pose significant health care problems, drug companies seldom develop drugs for orphan diseases due to extremely limited individual markets. Computer-aided drug repositioning is a cheaper and faster alternative to traditional drug discovery offering a promising venue for orphan drug research. We developed eRepo-ORP, a comprehensive resource constructed by a large-scale repositioning of existing drugs to orphan diseases with eThread, eFindSite and eMatchSite. A systematic exploration of 320,856 possible links between known drugs in DrugBank and orphan proteins obtained from Orphanet revealed as many as 18,145 candidates for repurposing.

Structural and computational biologists often need to measure the similarity of ligand binding conformations. The commonly used root-mean-square deviation (RMSD) is not only ligand-size dependent, but also may fail to capture biologically meaningful binding features. To address these issues, we developed the Contact Mode Score (CMS), a new metric to assess the conformational similarity based on intermolecular protein-ligand contacts. The CMS is less dependent on the ligand size and has the ability to include flexible receptors. In order to effectively compare binding poses of non-identical ligands bound to different proteins, we further developed the eXtended Contact Mode Score (XCMS). CMS and XCMS provide a meaningful assessment of the similarity of ligand binding conformations.

Molecular docking is an important component of computer-aided drug discovery. As part of a collaborative effort within LA-SiGMA, we developed GeauxDock, a new docking approach that builds upon 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. In order to drive docking simulations towards experimental conformations, the scoring function was carefully optimized to produce a correlation between the total pseudo-energy and the native-likeness of binding poses.

Alphaherpesviruses are a subfamily of herpesviruses that include the significant human pathogens herpes simplex viruses (HSV). HSV-1 glycoprotein K (gK) is a multi-membrane spanning virion glycoprotein essential for virus entry into neuronal axons, virion assembly, and pathogenesis. However, little is known about which gK domains and residues are most important for maintaining these functions across all alphaherpesviruses. We collaborate with Gus Kousoulas to elucidate the important structural features of gK that are involved in gK-mediated regulation of virus-induced membrane fusion. A greater understanding of mechanisms governing alphaherpesvirus membrane fusion is expected to inform the rational design of therapeutic and prevention strategies to combat herpesviral infection and pathogenesis.

Molecular docking has profound applications in drug discovery and development. 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. We proposed a new procedure for calculating the optimal docking box size that maximizes the accuracy of binding pose prediction and yields an improved ranking in virtual screening. Importantly, the optimized search space systematically gives better results than the default method not only for experimental pockets, but also for those predicted from protein structures. Our approach can be employed to fully automate large-scale virtual screening calculations by customizing docking protocols on the fly for individual library compounds.

The trend in library design for virtual screening has shifted to produce screening collections specifically tailored to modulate the function of a particular target or a protein family. In collaboration with Supratik Mukhopadhyay, we developed eMolFrag, a new software to decompose organic compounds into non-redundant fragments retaining molecular connectivity information. These building blocks can subsequently be employed by eSynth, an exhaustive graph-based search algorithm to computationally synthesize new compounds. Based on a small set of already developed bioactives, this approach is capable of generating diverse collections of molecules with the desired activity profiles.

The structural information on proteins in their ligand-bound conformational state is invaluable for protein function studies and rational drug design. However, the repertoire of the experimentally determined structures of holo-proteins is not only limited, but also these structures do not always include pharmacologically relevant compounds at their binding sites. To complement the existing repositories, we created eModel-BDB, a database of 200,005 comparative models of drug-bound proteins based on interaction data obtained from the Binding Database. Furthermore, we collaborate with Lukasz Kurgan to provide the Protein-Drug Interaction Database comprising a large number of putative protein-drug interactions that cover the entire structural human proteome.

The identification of protein-protein interactions (PPIs) is vital for understanding protein function, elucidating interaction mechanisms, and for practical applications in drug discovery. To improve the state-of-the-art in PPI modeling, we developed eFindSite^PPI to predict binding residues in a target protein structure with machine learning, and eRank^PPI to identify near-native conformations generated by protein docking with a new scoring function utilizing interface probability estimates and a contact-based symmetry score. Furthermore, we devised a high-throughput protocol for the bottom-up assembly of protein interaction networks based on all-to-all protein docking. These tools can be used to reliably identify and model biologically relevant protein assemblies at the proteome scale.

In higher plant Photosystem II, the PsbO, PsbP and PsbQ proteins provide critical support for oxygen evolution at physiological calcium and chloride concentrations. We collaborate with Terry Bricker to examine the structure of these components when bound to Photosystem II. This interdisciplinary project combines protein crosslinking, radiolytic footprinting coupled with high resolution tandem mass spectrometry, and molecular modeling to provide structural and functional information regarding the organization of the higher plant photosystem.

Detecting similarities between ligand binding sites in the absence of global homology between target proteins has been recognized as one of the critical components of modern drug discovery. Towards this goal, we developed eMatchSite, a new method to construct sequence order-independent alignments of ligand-binding sites with the Hungarian algorithm and machine learning. eMatchSite not only outperforms other approaches to match binding sites, but it also offers a remarkably high tolerance to structure distortions in protein models. Constructing biologically correct alignments opens up the possibility to investigate drug-protein interaction networks for complete proteomes with prospective systems-level applications in polypharmacology and rational drug repositioning.

As the frequency of antibiotic resistant bacteria steadily increases, there is an urgent need for new antibacterial agents. Because fatty acid synthesis is only used for membrane biogenesis in bacteria, the enzymes in this pathway are attractive targets for antibacterial development. Amino-oxazole inhibits biotin carboxylase (BC) activity in Gram-negative organisms. In collaboration with Grover Waldrop, we redesigned previously identified lead inhibitors of BC to expand the spectrum of bacteria sensitive to the amino-oxazole derivatives by including Gram-positive species. Structural insights into drug-BC interactions will be exploited to increase the potency of amino-oxazole inhibitors towards both Gram-negative as well as Gram-positive species.

A common strategy for virtual screening considers a systematic docking of a large library of organic compounds into the target sites in protein receptors. We developed eSimDock, a new approach to ligand docking and binding affinity prediction.  This algorithm employs non-linear machine learning-based scoring functions to improve the accuracy of ligand ranking, and similarity-based binding pose prediction to increase the tolerance to structural imperfections in the target structures. Comprehensive benchmarking calculations demonstrated that the performance of eSimDock is largely unaffected by the deformations of ligand binding regions, thus it represents a practical strategy for across-proteome virtual screening using protein models.

A growing body of evidence shows that gene products encoded by short open reading frames play key roles in numerous cellular processes. Yet, they are generally overlooked in genome assembly, escaping annotation because small protein-coding genes are difficult to predict computationally. There are still a considerable number of small proteins whose functions are yet to be characterized. To address this issue, we applied a collection of structural bioinformatics algorithms to infer molecular function of putative small proteins from the mouse proteome. Our results strongly indicate that many small proteins adopt three-dimensional structures and are fully functional, playing important roles in transcriptional regulation, cell signaling and metabolism.

Much needed functional annotation of gene products across different species often requires the knowledge of protein-ligand interactions. Towards this goal, we developed eFindSite, an algorithm to efficiently identify ligand-binding sites and residues from weakly homologous templates with highly sensitive meta-threading, improved clustering techniques, and advanced machine learning methods. eFindSite can also be used to conduct ligand-based virtual screening employing consensus molecular fingerprints. Carefully calibrated confidence estimates strongly indicate that highly reliable ligand binding predictions are made for the majority of gene products in a given proteome, thus eFindSite holds a significant promise for large-scale genome annotation and drug development projects.

Drug development is routinely streamlined using computational approaches to improve hit identification and lead selection, enhance bioavailability, and reduce toxicity. New challenges arose because processing a large volume of data demands unprecedented computing resources. As part of a collaborative effort within LA-SiGMA, we ported several of our codes to heterogeneous computing platforms. For instance, GeauxDock can be deployed on multi-core Central Processing Units (CPUs) as well as massively parallel accelerators, Intel Xeon Phi and NVIDIA Graphics Processing Unit (GPU). Further, a parallel version of eFindSite was implemented mainly for the Intel Xeon Phi platform. These parallel codes yield significant performance improvements considerably accelerating their large-scale applications.

Many structural bioinformatics approaches employ sequence profile-based threading algorithms. To improve fold recognition rates, homology searching may include artificially evolved amino acid sequences. We developed eVolver, an optimization engine that evolves protein sequences to stabilize the respective structures by a variety of potentials. Sequences generated by eVolver have high capabilities to recognize the correct protein structures using standard sequence profile-based techniques, thus they can be incorporated into existing threading approaches to increase their sensitivity. These sequences also provide a linkage between protein structure and function in computer simulations focused on the study of the completeness of protein structure space.

Template-based modeling that employs various meta-threading techniques is currently the most accurate, and consequently the most commonly used, approach to predict protein structures. We developed eThread, a highly accurate meta-threading procedure to identify structural templates followed by the construction of the corresponding target-template alignments and 3D models. We also extended the functionality of eThread to select functional templates covering a broad spectrum of protein molecular function, including ligand, metal, inorganic cluster, protein and nucleic acid binding. We demonstrated that in addition to accurate protein structure modeling, meta-threading effectively detects many facets of molecular function, even in a low sequence identity regime.