Spring 2013 Newsletter
Aaron Van Dyne
Aaron Van Dyne, (mathematics, physics) is now working with Dr. Dave Mathews at the University of Rochester Medical Center in computational biophysics. He is working on a MonteCarlo algorithm for predicting the secondary structure of RNA from the nucleotide sequence. The algorithm generates possible low free energy structures by performing millions of steps where a pair is formed or broken at each step. The structure is read out to a file every so often, and the file of structures represents possible low free energy structures, which could be the actual secondary structure. The idea of the algorithm, however, is to generate a variety of low free energy structures, so that the pairing probabilities can be estimated by counting how many of the structures have a particular pairing. The variability of the structure is enhanced by a process called replica exchange where the algorithm is run for multiple structures with temperatures between physiological temperature, 310.15K, and some higher temperature. The structures can then be exchanged from a higher temperature to a lower temperature at regular intervals. The exchange does not always occur, but it always occurs if the higher temperature structure has a lower free energy, and it occurs only with some probability if the higher temperature structure has a higher free energy. The larger the difference in free energy the less likely the exchange is to occur.
Aaron has done a lot of the work of debugging and testing the code. The algorithm and replica exchange are both functioning correctly now. The algorithm with the replica exchange has successfully predicted the known structure of two different 74 nucleotide tRNAs as one of the possible structures. The match between the pairing probabilities predicted by the algorithm and the pairing probabilities found using the partition function is currently being investigated. We are also currently investigating the optimal parameters of the simulation including getting good exchange probabilities. In the future, the algorithm will be extended to longer sequences and eventually sequences with a structural feature known as a pseudoknot.


Articles 


Links 


Faculty 

