We have developed and continue to improve upon an RNA folding algorithm (MPGAfold) that uses concepts from genetic algorithms. A recent version has been optimized and adapted to run on LINUX clusters, using MPI, such as a 256 processor SGI ALTIX with its high speed interconnect or on less expensive parallel PC-based architectures (including the new dual core PCs). The algorithm scales extremely well and is capable of running with hundreds of thousands of population elements on hundreds of physical processors, giving significant structural results when analyzed in the context of population variation. We are able to predict RNA pseudoknots and explore folding pathways that contain multiple functional conformations. In addition, the algorithm contains other features such as a Boltzmann relaxation technique, a choice of different energy rules, the ability to simulate sequential folding as well as sequential processing, forced/suggested and inhibited embedding of helical stems and the visualization of folding dynamics in real time. A new Java-based visualizer for depicting population evolution has also been developed which when coupled with the MPI version of MPGAfold makes the system more user friendly and portable and allow for a detailed exploration of the structure population space. STRUCTURELAB, the heterogeneous bioinformatical RNA analysis workbench, which permits the use of a broad array of approaches for RNA structure analysis, has been continually enhanced. 3D RNA structures can be generated and methodologies enable the comparison and analysis of multiple sequence RNA folds from a phylogenetic point of view, thus allowing improvement of predicted structural results across a family of sequences. In addition, visualization of folding pathways can be accomplished when StructureLab is used in conjunction with the genetic algorithm (see above). It is also possible to produce motif patterns so that families of RNA sequences can be explored for common structural elements. In general, STRUCTURELAB and other new tools we have developed, contain several features which when used together, act as set of data mining tools to aid in the discovery of patterns in databases of RNA structures. These systems have been adapted to other environments inside and outside our laboratory and NIH and are available upon request. We have also developed a new algorithm (""""""""KNetFold"""""""") for RNA secondary structure prediction. The methodology integrates thermodynamic and compensatory base change information using an innovative machine-learning algorithm (a hierarchical network of k-nearest neighbor classifiers). KnetFold has been shown to outperform other RNA secondary structure prediction programs. An another program CorreLogo has also been developed which depicts in a 3-dimensional plot correlations that exist between base pairs in a secondary structures. This new methodology uses mutual information derived from a sequence alignment. Both KnetFold and CorreLogo can be found as web servers on our website www.lecb/bshapiro/index.html. These systems have been employed in studying RNA structural characteristics, folding pathways and functional intermediates as exemplified by analysis of the folding pathways of the HIV 5' and 3' non-coding regions, as well as the hepatitis delta virus, interlukin-2, rotavirus and the turnip crinkle virus. In order to understand RNA structures, nanostructures, folding pathways and the structural effects of RNA-Protein interactions at the atomic level, structural elements of RNA molecules are being studied using molecular mechanics and molecular dynamics simulations. Studies include for example RNA tetraloops, bulge loops, kissing loops and three-way junctions. These studies have lead to the understanding of subtle atomic level interactions that will ultimately be quite significant to RNA molecular function and RNA nano design.
Showing the most recent 10 out of 16 publications
