Constraining Protein Structure Simulation Using Robotic Techniques with Aggregate Data

Document Type



Dr. Hans Dulimarta, dulimarh@gvsu.edu

Embargo Period



Molecular biology’s biggest challenge today is finding a way to accurately and quickly predict protein structures, since a protein’s structure determines its function. Computer simulations using both comparative approaches, which attempt to simulate an unknown structure based on how well its amino acid sequence matches the sequences of known structures, and structural approaches, which attempt to determine the structure from its amino acid sequence alone, have shown some promise. An intriguing variation of the structural approach uses robotic motion planning techniques, specifically random path planning, in trying to create a faster and more accurate model. The key insight of this approach is in realizing that a protein can be viewed as a semi-rigid chain with two degrees of freedom in each link, similar to a robot arm with multiple degrees of freedom. The chief drawback of the path planning approach is that it presupposes that one know the native state’s structure up front, as the roadmap is a series of randomly generated nodes normally distributed around the native state. In the experiment described in this paper, we experimented with different methods of creating a roadmap or configuration space by using some elements of both comparative and structural approaches. Specifically, we looked at using some of the available data on protein structure in a novel way, using linear regression to predict the range of sizes proteins with a certain number of residues would have. These allowed us to bound the configuration space and create a roadmap focusing on the final stages of the folding process, whose performance could be compared to the traditional roadmap focused on the native state.

This document is currently not available here.