Date of Award
Master of Science
Department of Systems Engineering and Management
Willie F. Harper, PhD.
The natural environment is burdened with a broad range of toxic chemicals, and there is a need to develop a tool that can accelerate the pace at which we learn how chemicals impact disease. This work developed an artificial neural network (ANN) based model that constructed chemical-disease relationships for chemicals found in the Comparative Toxicogenomics Database. A new chemical classification system, based on the molecular weight, hydrogen donors, and hydrogen acceptors, was created to identify chemicals with a unique number that is directly related to these structural properties of the chemical. Diseases were grouped into 27 categories and the chemical-disease associations were made between the chemical and its associated disease category. The ANN model was successfully trained and tested to associated 75 chemical with the 27 disease categories. Simulations with training-validation-testing ratios of 70-15-15 percent produced coefficients of determination equal to 0.99, and the Levenberg-Marquardt backpropagation function provided the best network performance. To help validate the model, the ANN was also used to evaluate chemical-disease relationships for three uncurated chemicals. Results showed that ANNs have the potential to predict disease associations for uncurated chemicals and to guide research for curated chemicals that may require further toxicological testing.
DTIC Accession Number
Brouch, Edward J., "Artificial Neural Network Prediction of Chemical-Disease Relationships Using Readily Available Chemical Properties" (2014). Theses and Dissertations. 702.