A Strategy for Improving the Performance of an SVMBased Protein Structure Scoring Algorithm

Publication Date

Fall 2017

Degree Type

Master's Project

Degree Name

Master of Science (MS)


Computer Science


The result of efforts to improve the state-of-the-art in protein structure scoring by developing a computational strategy to dynamically measure and consider local chemical properties of protein domains is presented. This was accomplished by examining the RFMQA and SVMQA protein scoring algorithms to understand their strengths and limitations with respect to scoring the structure of a set of CASP 8 through 12 target protein domain structures. Both of the RFMQA and SVMQA algorithms use, in part, several physical chemistry energy calculation methods to help converge on a score for a target protein target. It was observed that the conformation and local properties of a protein domain were not considered when obtain energy calculations. This resulted in non-optimal energy values being used to derive an overall score for the protein target. A computational strategy to modulate energy calculations based on properties of the protein domain being measured were developed to help improve scoring performance. A design of experiment (DOE) was developed to evaluate the strategy for SVMQA, but time and resources limitations precluded completing experiments. Completing the developed DOE successfully is expected to demonstrate at least a 10% improvement in scoring performance by the SVMQA algorithm. Having a purposeful protein structure scoring technology will help close the gap between the number of proteins being discovered and an ability to characterize their st5ructure.

This document is currently not available here.