In Silico Chemical Toxicity Prediction Models And The Prediction Of Risk Of Tobacco And Tobacco Smoke Components

Safety issues are recognized as an important cause of drug candidate failure nowadays. There is a strong requirement to bring safety assessment closer to earlier stage of the drug discovery process. Usually, traditional safety assessment methods require a large investment of time and money. Moreover, there is pressure to develop alternative approaches to toxicity testing that could reduce, refine or replace the use of animals, namely 3R principle, in safety assessment. In silico screening is typically a low cost high-throughput process, which can provide a fast indication of potential hazards for assessment of a large number of chemicals. In addition, these screens can be run on virtual compounds at early stages of drug discovery. Therefore, in silico toxicology prediction has become an essential tool in the research of chemical toxicology.In this thesis, we focused on the chemical toxicty prediction and the study of risk of tobacco and tobacco smoke components.In chapter 1, we gave a brief introduction of the background and several related methodoligies in computational toxicology, including read-across, qualitative/quantitative structure-toxicity relationships (QSTR) and structural alerts. Besides, we also described the method for meta-analysis.Chemical acute oral toxicity is an important endpoint in drug design and environmental risk assessment. However, it is difficult to be determined by experiments, and in silico methods are hence developed as an alternative. In chapter 2, a comprehensive data set containing 12,204 diverse compounds with median lethal dose (LD50) was compiled. These chemicals were classified into four categories based on the criterion of US Environmental Protection Agency (EPA). Then several multi-classification models were developed using five machine learning methods, including support vector machine (SVM), C4.5 decision tree (C4.5), random forest (RF), K-nearest neighbor (kNN), and naive Bayes (NB) algorithms, along with MACCS and FP4 fingerprints. One-against-one (OAO) and binary tree (BT) strategies were employed for SVM multi-classification. Performances were measured by two external validation sets containing 1678 and 375 chemicals, separately. The overall accuracy of MACCS-SVMOAOa model was 83.0% and 89.9% for external validation sets I and II, respectively, which showed reliable predictive accuracy for each class. In addition, some representative substructures responsible for acute oral toxicity were identified using information gain and substructure frequency analysis methods, which might be very helpful for further study to avoid the toxicity.Carcinogenicity is another toxicity endpoint that is widely concerned in human health, thus it is important to identify chemical carcinogenicity as early as possible. In chapter 3,829 diverse compounds with rat carcinogenicity were collected from Carcinogenic Potency Database (CPDB). Using six types of fingerprints to represent the molecules,30 binary and ternary classification models were generated to predict chemical carcinogenicity by five machine learning methods. The models were evaluated by an external validation set containing 87 chemicals from ISSCAN database. The best binary model was developed by MACCS keys and kNN algorithm with predictive accuracy at 83.91%, while the best ternary model was also generated by MACCS keys and kNN algorithm with overall accuracy at 80.46%. The results indicated that our models would be helpful for prediction of chemical carcinogenicity.Tobacco and tobacco smoke are probably the most significant sources of toxic chemical exposure and chemically mediated diseases in humans. Obviously, it will be very useful to provide the essential information of all the chemical components of tobacco and tobacco smoke for risk assessment research and the effective regulation of the hazardous products. In chapter 4, a database on chemical components of tobacco and tobacco smoke (CCTTS) was created. The database contains approximately 6000 chemical compounds identified in tobacco and tobacco smoke. For each compound, the chemical name, CAS Registry Number, structure, chemical type, physical and chemical properties, identification, and toxic activities were given. Among the chemical components,568 components with experimentally determined toxicities were collected from data mining, while 145 potential toxic components were predicted with above-developed prediction models. In addition, a user-friendly functional web interface was generated for browsing or searching the database, which would be available publicly soon.Many studies have reported an association between cigarette smoking and the risk of Parkinson’s disease (PD). In chapter 5, a meta-analysis was performed on published epidemiological studies to summarize the available information and evaluate the risk of PD associated with smoking. By searching publications in English language from 1960 to Oct. 2014, we obtained 61 case-control and 8 cohort studies that reported the risk estimates (odds ratio or relative risk) of Parkinson’s disease by cigarette smoking status. The pooled relative risk (RR) of Parkinson’s disease was 0.59 (95% CI,0.56-0.62) for ever smokers, which indicated that the risk of PD was 41% lower in ever smokers compared with never smokers. Several subgroup analyses were further carried out. All the results demonstrated the inverse association between cigarette smoking and the risk of PD. A significant inverse dose-response relationship was also observed for pack-years smoked. We hence suggest that effective drugs for PD might be developed using chemical substances derived from tobacco or tobacco smoke.In the last chapter, the whole work was summarized, and innovation points were emphasized.