| Popis: |
This thesis is concerned with the development and improvement of in silico toxicological models and will capitalise on the latest advances in predictive toxicological models at all levels of the toxicological QSAR pipeline. A review of data from in vitro and in vivo assays used for the prediction of mutagenicity and carcinogenicity is presented, collating over 10,000 molecules. The results of this review supported the subsequent development of an in silico model to predict the in vivo carcinogenicity assay. The overall model found 69.3% accuracy and 0.700 ROC AUC. This work was followed with an investigation into the use of quantum mechanical methods to predict lipophilicity (LogP). Solvated free energies in water and in 1-octanol were calculated using the M06-2X hybrid density functional and the def2-SVP basis set. The resulting model performed well in the SAMPL6 LogP Prediction Challenge where the model was placed seventh overall and notably superior to conventional methods. Finally, a larger investigation was conducted with skin sensitisation as the prediction target. Skin sensitisation is a toxicological outcome with scarce data available for the assays used in its prediction. The use of quantum mechanical calculations in this study enabled direct quantitative characterisation of electronic effects highly relevant to the skin sensitisation adverse outcome pathway. Ames mutagenicity models were also used for predicting skin sensitisation due to the importance of electrophilicity in both mechanisms of toxicity. Implicit solvation was incorporated into quantum molecular descriptor calculations as it was relevant to skin sensitisation mechanisms. The predictive performance achieved in the study was superior to that produced by the in vitro local lymph node assays for predicting human outcomes. These studies have generated insight into in silico chemical toxicity prediction methods and validate the use of state of the art computational methodologies. |
