Existing structure-prediction software packages have inconvenient limitations. For instance, MFOLD folds sequences only one at a time, and the Vienna package uses only thermodynamics and mutual information to weight its structure predictions. However, multiple sequence alignments include additional data that can be used to calculate measures (such as covariation values and mismatch data) capable of increasing the accuracy of structure predictions. This makes existing software a poor match for the needs of researchers whose work produces multiple sequence alignments.

BayesFold, which is designed to accept an entire alignment as input instead of a single sequence, makes use of the alignment's additional information to weight its structures. It relies on the Bayesian method to weight these various types of data in a statistically sound way that takes the guesswork out of combining multiple probabilities. In addition, BayesFold offers significant interface optimizations for viewing multiple sequences threaded through the returned structures.