View Folding Results
Next: Methods Up: Discussion Previous: Merits of annotation
The ``Vienna group'' has developed several additional terms, based on statistical mechanics and information theory, that can be used to describe how well determined a predicted structure or sub-section thereof is. These include a ``well-definedness'' measure, d(i) and an entropy measure, S(i), for every base i in the RNA [Huynen et al., 1997]. These descriptors can also be used to annotate predicted structures with color in the manner that we have described. These descriptors are neither better nor worse than the base pair probabilities but they convey different information. It still remains to be seen whether these alternative approaches will provide additional insight into RNA structure that would be useful to the experimentalist.
Although base annotation based on P-num has already proved extremely useful in several analyses of RNA structure [Jacobson et al., 1998,Palmenberg & Sgro, 1998], the approach relies heavily on the basic reliability of the RNA structure prediction itself and is therefore very much dependant on the energy rules that are used in RNA structure prediction. Although the prediction of local hairpins now appears to be quite reliable, multi-branch junctions and long range interactions remain problematic. Work is currently underway to implement the helix stacking feature in multibranch junctions that is described in [Walter et al., 1994]. It is expected that the implementation of this feature will improve the reliability of long range prediction. In addition, it may alter the ``well-definedness'' of individual structural features.
Institute for Biomedical Computing
Washington University in St. Louis
August 21 1998.