With the recent explosion of new methods for fold recognition, an objective way to evaluate their performance is needed. We have developed a benchmark to assess the performance of fold-recognition methods. The benchmark consists of a set of 68 protein sequences matched by superposition to known structures. This set covers a wide range of protein families, and includes matching proteins with insignificant sequence similarity.
The user of the benchmark evaluates his/her own method by searching the benchmark's library of 301 folds of known structure, with each of the 68 test sequences , in order to find the most compatible fold. The performance is evaluated using two criteria: sensitivity and selectivity.