LEARNING TO FOLD RNA WITH PARALLEL PROCESSORS

We implement a parallel processed simulation of sequential folding pathways occurring concurrently as a growing RNA chain explores the space of available secondary structures. An appropriate learning or updating strategy is introduced to systematically trim the branching of sequential folding pathways. By applying our algorithm to specific RNA sequences with known native structure, we provide evidence that this learning strategy determines the expeditious convergence to the actual ensemble of optimal and suboptimal secondary structures. The robustness of our results to uncertainties in the thermodynamic parameters is shown to result from the very nature of the parallel process. Our final goal is to examine the dependence of the learning efficiency on the information content of the primary sequence. This is done by contrasting the folding process of specific RNA species whose native structures are known with the folding of random sequences.