Evaluation of inter- and intramolecular primary structure homologies of interferons by a Monte Carlo method

Using Sellers TT algorithm, primary structure repeats have been described for interferon (IFN)-alpha, -beta 1, and gamma. To reevaluate these results and to extend them to IFN-beta 2 (interleukin-6), a modified algorithm was developed that uses a metric to define the "best" partial homology of two peptide sequences and to compare it to those detected in random permutations of the peptide. Using this approach, the known structural homologies of IFN-alpha with IFN-beta 1 and of human (Hu) IFN-gamma with murine (Mu) IFN-gamma were identified correctly. However, the primary structure repeats in the amino acid sequences of IFN-alpha, -beta 1, and -gamma turned out to be no better than those detectable in random permutations of these sequences. These results were confirmed using a different, nonlinear metric. A previously used approach to demonstrate significance was shown to produce false-positive results. No significant primary structure homologies were detected among IFN-beta 1, -beta 2, and -gamma. In contrast to the amino acid sequence analysis, the DNA sequence of HuIFN-beta 1 contained a significant repeat that had no significant counterpart in MuIFN-beta or in IFN-alpha. In conclusion, some previously reported results obtained with Sellers TT algorithm on amino acid sequences are easily explained as random similarities, and it is therefore strongly recommended that a method like ours should be used to control significance.