How universal are fold recognition parameters. A comprehensive study of alignment and scoring function parameters...

A choice of sequence-structure similarity scoring function parameters can significantly alter results of the performance of the recognition of distantly related folds. It therefore constitutes a critical part of fold recognition process. In order to increase an understanding of the influence of parameter choice, a comprehensive benchmark of very hard (SFOLD) and medium hard (SFAM) fold recognition examples has been derived from the SCOP database of protein structure families. These benchmarks have subsequently been used to optimize, validate and analyze dependence of recognition sensitivity on alignment and fold similarity score parameters for different scoring functions. Significant variation of the common parameters has been observed for different functions, leading to the conclusion that optimal parameter sets are not universal. The scope of solutions common to any pair of scoring function is relatively small, hence, using jury method for fold prediction seems not appropriate. Also, using a redundant version of fold libraries significantly increases odds of identification of distantly related fold.