NovaFold Benchmarks

DNASTAR NovaFold was evaluated by recreating the CASP11 competition, in which participants attempted to predict the structures for a given set of target sequences whose structures were previously unknown. In this benchmark test, NovaFold created predictions for these same target sequences which were then evaluated and scored based on the same metrics used in the CASP11 experiment.

A normalized measure of similarity between superimposed protein structures weighs close atom pairs between structures greater than more distant atom pairs. The range varies from 0 to 1, with a higher score indicating a better structural match.

Additional Notes:

  • NovaFold versions 12.3 and higher support sequences up to 2,000 residues in length.
  • To accurately simulate the conditions of the original competition, the benchmark was run using a library composed only of templates available prior to the end date of the CASP11 contest. In addition, targets leaked during the competition were removed from the benchmark library.
  • This benchmark test utilized the Cloud-based version of NovaFold and evaluated structure prediction only, not binding or function prediction.

NovaFold Runtimes
NovaFold runtimes plotted by sequence length and colored by TM-score