Combinatorial Chemistry & High Throughput Screening - Volume 18, Issue 6, 2015

Advancement in chemoinformatics research in parallel with availability of high performance computing platform has made handling of large scale multi-dimensional scientific data for high throughput drug discovery easier. In this study we have explored publicly available molecular databases with the help of open-source based integrated in-house molecular informatics tools for virtual screening. The virtual screening literature for past decade has been extensively investigated and thoroughly analyzed to reveal interesting patterns with respect to the drug, target, scaffold and disease space. The review also focuses on the integrated chemoinformatics tools that are capable of harvesting chemical data from textual literature information and transform them into truly computable chemical structures, identification of unique fragments and scaffolds from a class of compounds, automatic generation of focused virtual libraries, computation of molecular descriptors for structure-activity relationship studies, application of conventional filters used in lead discovery along with in-house developed exhaustive PTC (Pharmacophore, Toxicophores and Chemophores) filters and machine learning tools for the design of potential disease specific inhibitors. A case study on kinase inhibitors is provided as an example.

In this work we present ChemScreener, a Java-based application to perform virtual library generation combined with virtual screening in a platform-independent distributed computing environment. ChemScreener comprises a scaffold identifier, a distinct scaffold extractor, an interactive virtual library generator as well as a virtual screening module for subsequently selecting putative bioactive molecules. The virtual libraries are annotated with chemophore-, pharmacophore- and toxicophore-based information for compound prioritization. The hits selected can then be further processed using QSAR, docking and other in silico approaches which can all be interfaced within the ChemScreener framework. As a sample application, in this work scaffold selectivity, diversity, connectivity and promiscuity towards six important therapeutic classes have been studied. In order to illustrate the computational power of the application, 55 scaffolds extracted from 161 anti-psychotic compounds were enumerated to produce a virtual library comprising 118 million compounds (17 GB) and annotated with chemophore, pharmacophore and toxicophore based features in a single step which would be non-trivial to perform with many standard software tools today on libraries of this size.

NMR based chemical shifts are an important diagnostic parameter for structure elucidation as they capture rich information related to conformational, electronic and stereochemical arrangement of functional groups in a molecule which is responsible for its activity towards any biological target. The present work discusses the importance of computing NMR chemical shifts from molecular structures. The NMR chemical shift data (experimental or computed) was used to generate fingerprints in binary formats for mapping molecular fragments (as descriptors) and correlating with the bioactivity classes. For this study, chemical shift data derived binary fingerprints were computed for 149 classes and 4800 bioactive molecules. The sensitivity and selectivity of fingerprints in discriminating molecules belonging to different therapeutic categories was assessed using a LibSVM based classifier. An accuracy of 82% for proton and 94% for carbon NMR fingerprints were obtained for anti-psoriatic and anti-psychotic molecules demonstrating the effectiveness of this approach for virtual screening.

The target ligand association data is a rich source of information which is not exploited enough for drug design efforts in virtual screening. A java based open-source toolkit for Protein Ligand Network Extraction (J-ProLiNE) focused on protein-ligand complex analysis with several features integrated in a distributed computing network has been developed. Sequence alignment and similarity search components have been automated to yield local, global alignment scores along with similarity and distance scores. 10000 proteins with co-crystallized ligands from pdb and MOAD databases were extracted and analyzed for revealing relationships between targets, ligands and scaffolds. Through this analysis, we could generate a protein ligand network to identify the promiscuous and selective scaffolds for multiple classes of proteins targets. Using J-ProLiNE we created a 507 x 507 matrix of protein targets and native ligands belonging to six enzyme classes and analyzed the results to elucidate the protein-protein, protein-ligand and ligand-ligand interactions. In yet another application of the J-ProLiNE software, we were able to process kinase related information stored in US patents to construct disease-gene-ligand-scaffold networks. It is hoped that the studies presented here will enable target ligand knowledge based virtual screening for inhibitor design.

Virtual screening is an indispensable tool to cope with the massive amount of data being tossed by the high throughput omics technologies. With the objective of enhancing the automation capability of virtual screening process a robust portal termed MegaMiner has been built using the cloud computing platform wherein the user submits a text query and directly accesses the proposed lead molecules along with their drug-like, lead-like and docking scores. Textual chemical structural data representation is fraught with ambiguity in the absence of a global identifier. We have used a combination of statistical models, chemical dictionary and regular expression for building a disease specific dictionary. To demonstrate the effectiveness of this approach, a case study on malaria has been carried out in the present work. MegaMiner offered superior results compared to other text mining search engines, as established by F score analysis. A single query term 'malaria' in the portlet led to retrieval of related PubMed records, protein classes, drug classes and 8000 scaffolds which were internally processed and filtered to suggest new molecules as potential anti-malarials. The results obtained were validated by docking the virtual molecules into relevant protein targets. It is hoped that MegaMiner will serve as an indispensable tool for not only identifying hidden relationships between various biological and chemical entities but also for building better corpus and ontologies.

The power of cloud computing and distributed computing has been harnessed to handle vast and heterogeneous data required to be processed in any virtual screening protocol. A cloud computing platorm ChemInfoCloud was built and integrated with several chemoinformatics and bioinformatics tools. The robust engine performs the core chemoinformatics tasks of lead generation, lead optimisation and property prediction in a fast and efficient manner. It has also been provided with some of the bioinformatics functionalities including sequence alignment, active site pose prediction and protein ligand docking. Text mining, NMR chemical shift (1H, 13C) prediction and reaction fingerprint generation modules for efficient lead discovery are also implemented in this platform. We have developed an integrated problem solving cloud environment for virtual screening studies that also provides workflow management, better usability and interaction with end users using container based virtualization, OpenVz.