Date of Award

Summer 2005

Degree Type

Thesis - Restricted

Degree Name

Master of Science (MS)


Mathematics, Statistics and Computer Science

First Advisor

Sem, Daniel

Second Advisor

Struble, Craig A.

Third Advisor

Povinelli, Richard J.


One aspect of drug design involves filtering libraries of existing compounds in order to select those that interact strongly with the protein of interest. An active area of drug design research is creating computational filters for this process. This work develops a machine learning approach for constructing a prediction model for the interactions of drugs with Cytochrome P450 2D6 (CYP2D6). This enzyme is responsible for metabolizing 20-25% of all clinically utilized drugs. To construct this model, a set of attributes are created for each drug compound. The novelty of this approach is the use of results from molecular docking simulations as attributes. Furthermore, these compounds are docked into multiple low energy conformations of CYP2D6 that are distinct in 3-dimensional space. The final model performed with a 78% accuracy when predicting whether a given compound is a binder, non-binder, or has a moderate interaction with CYP2D6, which is comparable with current machine learning approaches for this type of problem.