Computational RNomics

Structural genomics, the systematic determination of all macro-molecular structures represented in a genome, is at present focused almost exclusively on proteins. Although it is commonplace to speak of ``genes and their encoded protein products``, thousands of human genes produce transcripts that exert their function without ever producing proteins. Furthermore, even though the sequence of the human DNA is known by now, the contents of about half of it remains unknown. It is quite likely that a large class of genes has gone relatively undetected so far because they do not make proteins. The list of functional non-coding RNAs includes key players in the biochemistry of the cell, such as transfer RNAs, ribosomal RNAs, tmRNA, and the RNA components of RNase P and signal recoginition particles. Another level of RNA function is presented by functional motifs within protein-coding genes, located mostly in the non- translated 5` or 3` regions of the immature messenger RNA. It is not hard to argue therefore that "RNomics", i.e., the understanding of functional RNAs and their interactions at a genomic level, is of utmost practical and theoretical importance in modern life sciences: The comprehensive understanding of the biology of a cell obviously requires the knowledge of identity of all encoded RNAs, the molecules with which they interact, and the molecular structures of these complexes. The first step toward this goal is the development of versatile and reliable computational methods that can detect and classify functional RNAs, preferably within a single genome, or in case this proves impossible, from a very small set of related genomes. We propose here to develop a suite of bioinformatics methods that are specifically geared toward detecting, verifying, and classifying functional RNAs. Our comprehensive approach to "Computational RNomics" will provide * improved algorithms for RNA secondary structure prediction * improved alignment algorithms for nucleic acid sequences * novel approaches to compare and align RNA structures * extensions of existing RNA algorithms to deal with genome-size data sets * a database system specifically designed for RNA structures

RNomics stands for the holistic approach to look at all RNAs existing in living organisms. Prediction, detection and classification of these RNAs requires tools using comparative methods for identification of similarities between an unknown and a described RNA species. In the case of DNA, classification is performed simply by sequence comparison. Proteins in turn require additional steps since their function is determined by their three dimensional structure. The function of an RNA is represented by only two constraints: the sequence, which folds into a distinct secondary structure. Folding follows a distinct set of rules, only three kinds of base pairs are allowed: A-U, G-C, G-U. Programs like Mfold or RNAfold calculate such secondary structures starting from a single input sequence. When comparing an RNA from two different organisms, say snoRNA E1 in mouse and man, the sequence will be different while the secondary structure is more or less the same. This phenomenon is a result of compensatory mutations, which are changes in the sequence such that a base pair remains preserved. Only those base pairs taking part in a functional structure will be under such selection pressure. Thus, detection of compensatory mutations allows us to decide wether two ore more similar sequences belong to the same class of functional RNAs. Aim of this project was the development of a variety of tools supporting the quest for RNA genes. We were able to find solutions to all problems formulated in the grant proposal and apply them on selected model organisms of viruses, bacteria, und animals. Thus we were able to proof the general applicability of our RNA tools. We developed two algorithms for aligning RNA sequences taking into account protein coding capacity (codaln) and secondary structure information (pmmulti), respectively. The programs hxmatch and alidot use different methods to predict conserved structures, including computationally expensive pseudoknots, from alignments. Both Alifoldz and RNAz also follow the comparative approach and represent methods for genome wide surveys in eukaryote genomes. Handling of such huge amounts of data requires a specifically designed MySQL database. RNALfold provides a sliding window variant of RNAfold for prediction of locally stable RNA secondary structures and is also applicable on whole genomes. In addition to the proposed goals, we developed methods for designing bistable RNAs, so called RNAswitches and small modifier RNAs, which inhibit formation of functional mRNA structures. The molecular evolution of a group of microRNAs, a natural class of small non- coding RNAs, was reconstructed based on sequences found by means of tools developed during this project.