Protein folds from pair interactions: a blind test in fold recognition.

We submitted nine predictions to CASP2 using our fold recognition program ProFIT. Two of these structures were still unsolved by the end of the experiment, six had a recognizable fold, and one fold was new. Four predictions of the six recognizable folds were correct. Two models were excellent in terms of alignment quality (T0031, T0004): in one the alignment was partially correct (T0014), and one fold was correctly identified (T0038). We discuss improvements of the program and analyze the prediction results.