The current best algorithm for RNA secondary structure prediction achieves an accuracy of about 70% for pseudoknot-free secondary structures. Part of this weak performance is due to the underlying thermodynamic model, which is incomplete and/or inaccurate. The parameters used in this model have been determined by enormous work of some biochemistry labs over several years. In this talk I am going to propose a computational model to predict the RNA thermodynamic parameters. The problem is defined as follows: given a large set of RNA sequences with known MFE secondary structures, determine the best underlying thermodynamic parameters for RNA secondary structure prediction. I propose to use "Iterative scaling algorithm", a supervised learning algorithm that has been used successfully for parameter estimation for other exponential models.

So far, this is just an idea that I might explore in the future. It came to me from a lot of work with RNA secondary structure prediction, which sometimes was not quite accurate. Your opinion and suggestions will be much appreciated.