SignalP V1.1 World Wide Web Prediction Server Center for Biological Sequence Analysis

Background: Prediction of signal peptides

With the huge amount of unprocessed sequence data, automatic prediction of protein function and location becomes increasingly important. One important aspect of this is the prediction of secretory signal peptides and location of their cleavage sites.

The most widely used method is a weight matrix published by von Heijne in 1986 (Nucleic Acids Res., 14, 4683-4690). The method is still used with the original matrix data today, even though the amount of signal peptide data available has increased by a factor of 5-10 since 1986.

Here, we present a method based on a larger and more recent data set, and a more sophisticated computational tool: artificial neural networks. These have been used for many biological sequence analysis problems, including prediction of protein secondary structure and mRNA splice sites. In this approach, we use one network to distinguish between signal peptides and non-signal peptides and another network to recognize the cleavage site.