Date of Award

Fall 2006

Degree Type

Thesis - Restricted

Degree Name

Master of Science (MS)

Department

Mathematics, Statistics and Computer Science

First Advisor

Struble, Craig A.

Second Advisor

Sem, Daniel S.

Third Advisor

Povinelli, Richard J.

Abstract

Developing a new drug is a complex process. Today, with the use of combinatorial chemistry, millions of molecules are rapidly produced and tested by pharmaceutical research companies. Computer-based virtual screening methods are then used for pre-selecting molecules as potential binders of a target protein in the body for further investigation. The main idea behind this research is to build a virtual screening filter for Cytochrome P450 2D6 (CYP2D6) drugs using neural networks. CYP2D6 is of interest in the pharmaceutical industry because it metabolizes 25% of the clinically utilized drugs and eliminates them quickly from the body. Also, inhibition of metabolism of a CYP2D6 drug by another drug may cause harmful drug-drug interactions. We construct an artificial neural network model for predicting binding affinities of compounds with an accuracy of 85.36±6.5%. This model is an improvement of a previous model that performed with an accuracy of 78±6.5%. In building our new model, we explore the effect of two new features, druglikeness score and best docking features, on the performance of the model. We show that our model containing best docking features improved the prediction accuracy. We also demonstrate that druglikeness feature does not contribute towards performance. In order to understand the working of our neural network model, we extract rules that are used by it to classify the compounds as binders, moderates and non-binders. Rule extraction enabled us to hypothesize properties of binders and non-binders. We find that the rules we obtained for binders are consistent with literature. Additionally, we study the incorrectly classified compounds to identify the reasons for their misclassification and attempt to propose new features by structural analysis.

Share

COinS