Graph generating systems for predicting biological structures
Shareghi Arani, Pooya
MetadataShow full item record
Computational predictions of RNA and protein structures have become necessary complements to the expensive experimental methods of structure elucidation. In this dissertation, we present two novel approaches aimed at improving the efficiency and accuracy of structure prediction. Both approaches involve generating biological structure graphs; one directly generates optimal graphs, while the other does so indirectly by generating tree-decompositions of the optimal graphs. First, we discuss an algorithm for simultaneously predicting the RNA secondary structure and coaxial stacking of the predicted helices. The results show an additional 17% improvement over the state-of-the-art program RNAfold in accurately predicting tRNA structures. Second, we describe a framework to facilitate the efficient ab-initio prediction of RNA and protein tertiary structures. At the core of this framework is a context-sensitive set of rules that is used to generate tree-decompositions of optimal tertiary structure graphs. This context-sensitive system effectively limits the search space to improve both the running time and space.