Date of Award

Summer 8-2016

Degree Type

Honors College Thesis

Department

Computing

First Advisor

Zheng Wang

Advisor Department

Computing

Abstract

Protein residue-residue contact prediction is one of many areas of bioinformatics research that aims to assist researchers in the discovery of structural features of proteins. Predicting the existence of such structural features can provide a starting point for studying the tertiary structures of proteins. This has the potential to be useful in applications such as drug design where tertiary structure predictions may play an important role in approximating the interactions between drugs and their targets without expending the monetary resources necessary for preliminary experimentation. Here, four different methods involving deep learning, support vector machines (SVMs), and direct coupling analysis were trained on a dataset of proteins from the 9th Critical Assessment of Techniques for Protein Structure Prediction (CASP 9). The models that were the most successful after training on the CASP 9 data were selected to perform the contact predictions in each method. After performing a blind test on CASP 11 targets, we have determined that further optimizations to the training process may be necessary to improve performance.

Share

COinS