Date of Award


Document Type

Thesis (Master's)

Department or Program

Department of Computer Science

First Advisor

Scot Drysdale


RNA secondary structure prediction is an area where computational techniques have shown great promise. Most RNA secondary structure prediction algorithms use dynamic programming to compute a secondary structure with minimum free energy. Energy minimization algorithms are less accurate on larger RNA molecules. One potential reason is that larger RNA molecules do not fold instantaneously. Instead, several studies show that RNA molecules fold progressively during transcription. This process could encourage the molecule to fold into a structure that is not at the global lowest energy level. Additionally, dynamic programming algorithms do not allow for a important type of structure called a pseudoknot. Secondary structure prediction allowing pseudoknots was recently shown to be NP-complete. We have created a simulation that captures these biological insights. Our simulation uses a probabilistic approach to fold the molecule progressively as it is synthesized. This thesis evaluates the performance of the simulation and presents several enhancements to improve efficiency and accuracy. Our results show that our progressive folding algorithm did not improve on current techniques. Additionally, we found that a simulated annealing algorithm using our probability models was more accurate than our progressive folding algorithm.


Originally posted in the Dartmouth College Computer Science Technical Report Series, number TR2017-835.