next up previous contents
Next: Introduction Up: Algorithms and Thermodynamics for Previous: Contents


his article is about the current status of the mfold package for RNA and DNA secondary structure prediction using nearest neighbor thermodynamic rules. The details of the free energy rules and of the latest version 3.0 software are described. Future plans are also discussed.

The mfold software now runs on a variety of Unix platforms; specifically SGI Irix, Sun Solaris, Dec alpha Ultrix and on Linux. While the older interactive programs of version 2 still exist, they are now run by a variety of scripts that make for much easier handling. There is both a command line interface for mfold and an HTML interface that runs in a Unix environment but can be accessed by anyone with a web browser.

The thermodynamic basis for the folding model is presented in detail, with references given for both specific free energy parameters and to overview articles that have summarized the state of these nearest neighbor parameters over the past dozen years. Both RNA and DNA rules are discussed, with some mention of parameters for RNA/DNA hybridization. Although the thermodynamic model has grown in complexity to accommodate new types of information, the folding algorithm has not yet incorporated some features, such as coaxial stacking of adjacent helices, and other features will probably remain too difficult or computationally expensive to implement. For this reason, a new energy calculation program has been introduced to recompute the free energies of predicted foldings to reflect the best of our knowledge.

The most significant improvements in the mfold software are:

  1. Folding times have been greatly reduced in recent years, partly because of faster computers and partly because of improvements in the algorithm.
  2. Folding constraints have been expanded and are now implemented without the use of bonus energies that distort the results.
  3. The output is significantly improved. Clear and enlargeable images of dot plots and of predicted foldings are now produced in PostScript and gif formats. Bases in structures may be annotated using different colors that reflect how well-determined they are in terms of their tendency to pair with other bases or to be single-stranded. Similarly, base pair probabilities from partition function calculations may be used for annotation. A detailed decomposition of each predicted folding into stacked pairs and loops with associated free energies is now provided.

The mfold software has a variety of parameters that may be adjusted to improve the predictions. Several examples are presented to illustrate how to interpret folding results and how to adjust these parameters to obtain better results.

next up previous contents
Next: Introduction Up: Algorithms and Thermodynamics for Previous: Contents

Michael Zuker
Center for Computational Biology
Washington University in St. Louis