The Methodology Development And Applications Of Systems Drug Design

The traditional drug design paradigm based on the hypothesis of "one gene, one drug, one disease" is seeking to designing high active and selective ligands to act on individual drug targets in drug discovery. However, the traditional dominant paradigm was challenged by the high clinical attrition failure rate. For example, although drugs with high potency and selectivity ("magic bullet" drugs) showed ideal in vitro biological activities, they often generated some adverse side effects or toxicity in vivo. One main reason is the complexity of biological systems and the mechanism-of-action (MOA) of drugs in vivo. The process of a drug in organisms included absorption, distribution, metabolism and excretion (ADME), and MOA of a drug included the interactions with proteins and toxic profiles, which was called ADMET profiles and drug-target interaction (DTI) spectra. Under the force of these factors, there is a novel paradigm of systems drug design, recently. The systems drug design is to study the key issues in the area of drug design via the systematic or integrative methodologies, that is development of computational models or mathematical equations to explore the complex networks and pathways of drugs in single cell or whole organisms by integration of open data source, knowledge and technique of multidisciplinaries, including chemistry, biology, mathematics, informatics and computer science etc., to discovery ideal drugs with high potency and low toxicity for human health. This thesis is mainly focused on the development and application of new methodologies and tools for systems drug design, including two sections as follows.The first section is mainly focused on development of new computational methodologies and tools for prediction of DTI network, drug side effects and drug repositioning, in order to accelerate development and application of systems drug design in drug discovery. In chapter2, we developed drug-based similarity inference, target-based similarity inference, network-based inference (NBI) and weighted NBI for prediction of DTI and drug repositioning. Five old drugs, namely montelukast, diclofenac, simvastatin, ketoconazole, and itraconazole, were identified to have novel polypharmacological effects on human estrogen receptors or dipeptidyl peptidase-â…£ with half maximal inhibitory or effective concentration from submicromolar to micromolar by computational prediction and in vitro assays. In addition, the "weak-interaction hypothesis" in DTI network was found by the edge-weighted NBI method. In chapter3, we reported a comprehensive database of adverse drug events (abbreviated as MetaADEDB) by data integration and text mining. To demonstrate the usage of MetaADEDB. the human phenotypic network inference and drug side-effect similarity inference methods were developed for prediction of new potential side effect of drugs and DTI. High performance was yielded and the molecular mechanisms of unknown side effects of several old drugs were successfully elucidated. In chapter4. we reported the development and critical assessment of multi-target quantitative-structure activity relationship (mt-QSAR) and computational chemogenomics methods. Although the high performance were yielded for protein sequence-based computational chemogenomic method in the internal cross-validation, the further results revealed that there was a high false positive rate in the external validation set when using computational chemogenomic method. In addition, the software and web server named CPI-Predictor was developed for freely predicting target of new chemicals using the high predictive mt-QSAR models.In the second section, we developed several new methodologies, database and software for filtering/predicting ADMET profile of compounds in drug discovery and environmental risk assessment. In chapter5, we developed a new combined classifier algorithm based on the pattern recognition technique for prediction of cytochrome P450(CYP450) inhibitor and non-inhibitor. The results showed that the performance of our combined classifier outperform three classic methods, including Mean. Maximum and Multiply. Herein, we developed an entropy-based index to quantify and predict the CYP450inhibitory promiscuity of small molecules. The results indicated that the CYP450inhibitory promiscuity of compounds would have a moderate correlation with molecular aromaticity, a minor correlation with molecular lipophilicity, and no or very low correlation with molecular complexity, hydrogen bonding ability and TopoPSA. The new computational methodology and several high predictive classification models with probability output to predict chemical biodegradability was developed based a comprehensive chemical biodegradability database. The experimental assays demonstrated the good generalization ability of our models. In chapter6. several computational methodologies and predictive classification and regression models were developed for filtering toxicity on three important aquatic and terrestrial toxic organisms, including tetrahymena pyriformis, fathead minnow and honey bee. In addition, we also developed a novel computational systems toxicology framework for evaluating the ecotoxicological risk of chemical substances. Comparing with traditional QSAR methods, the computational systems toxicology framework not only-predicted new toxic phenotypes. but also predicted new chemical-gene or-protein interaction network at the same time, which should be more useful for chemical safety profiling. In chapter7. an ADMET structure-activity relationship server, abbreviated as admetSAR (www.admetexp.org). was reported using text mining technique. It is a comprehensive, open source, text and structure searchable database. In admetSAR, about 220,000ADMET annotated data points for more than96,000unique compounds with45kinds of ADMET-associated properties, proteins, species or organisms have been carefully curated from a large number of diverse literatures. In addition, the database includes22qualitative classification and5quantitative regression models with highly predictive accuracy, allowing to predicting or filtering ecological/mammalian ADMET properties of novel chemicals. The admetSAR could be very useful for drug design, drug development and environmental risk assessment, which has been successfully used for evaluating chemical biodegradability in case study. So far, admetSAR had been visited more than5000times, and is widely used by more than30domestic and foreign pharmaceutical companies and academic institutions.