David H. Mathews, M.D., Ph.D.

Our understanding of the role of RNA in cellular processes has expanded enormously over the last two decades. Originally, RNA was understood to participate in protein expression as a carrier of genetic information (mRNA) and as an adapter molecule (tRNA) for reading the code. Then RNA was discovered to catalyze reactions, including self-splicing, phosphodiester bond cleavage, and peptide bond formation. RNA is now known to play functions in diverse cellular processes, such as development, immunity, RNA editing and modification, and post-transcriptional gene regulation. RNA is also an important player in many diseases, including Prader-Willi, b-thalassemia, and myotonic dystrophy. RNA sequences can be evolved in vitro to catalyze many reactions that are not part of the natural repertoire. Antisense and RNAi can be used to modulate gene expression.

Research in the Mathews lab spans the fields of Computational Biology and Bioinformatics. We are interested in predicting RNA structure and we develop computational tools for targeting RNA with pharmaceuticals and for using RNA as a pharmaceutical (Mathews et al., 1999a).

In collaboration with Doug Turner (University of Rochester) and Michael Zuker (RPI), we have developed software that predicts secondary structure, i.e. the canonical base pairs (Mathews et al., 2004; Mathews et al., 1999b). On average, 73% of base pairs are correctly predicted in a set of diverse sequences with known structures. This accuracy can be improved by constraining the structure prediction using data derived from experiments.

We have also developed software that uses a partition function to predict base pairing probabilities (Mathews, 2004). Using this algorithm, secondary structures can be color annotated according to pairing probability to graphically demonstrate both high probability pairs and low probability pairs that are, on average, not as accurate.

Finally, we are developing methods to predict a secondary structure common to multiple sequences (Mathews & Turner, 2002). The accuracy of structure predictions is dramatically improved by using the information contained in multiple sequences. For example, for a set of poorly predicted 5S rRNA sequences, the average accuracy of base pair prediction improves from 47.8% to 86.4% when the structure common to two sequences is determined.