RNA Structures Space is Largely Vacant Except for a Small Number of Local Energy Wells | Department of Mathematics

RNA Structures Space is Largely Vacant Except for a Small Number of Local Energy Wells

Event Information
Event Location: 
GAB 461, 3-4PM; Refreshments: GAB 472, 2:30-3PM
Event Date: 
Monday, October 20, 2014 - 3:00pm

RNA secondary structure prediction often consists of choosing a candidate structure using the thermodynamic nearest neighbor model to find the lowest energy structure. The implicit assumption is that one structure dominates the structural landscape. We find that this assumption is false for most RNA secondary structures. Using an information theoretic criterion we show that: The Boltzmann weighted space of these structures is exemplified by a finite mixture of energy wells defined by the presence of conflicting basepairs; and a small number of these wells contain a high proportion of the Boltzmann probability mass but only a trifling fraction of the ensemble of secondary structures. We develop an information theoretic algorithm to identify these energy wells. The algorithm returns a mutually exclusive set of these wells and a fully probabilistic model for each well. Enumerating these alternate structures is important for structure prediction where a small change in the energy model could lead to a preference for one alternate structural well over another. This uncertainty largely disappears when chemical modification data is taken into account. Beyond structure prediction, finding local energy wells is relevant to RNA kinetics, where an RNA molecule must overcome the energy barriers between these wells as it folds.