Sorry I should have copied the users group on this: Yes that is a problem we have encountered. We have basically decided to use each part for what it is "advertised" to do best: we use the HMM result as being slightly more accurate for eukaryotic data "For eukaryotic data, SignalP-HMM has a substantially improved discrimination between signal peptides and uncleaved signal anchors" (quote taken from their site) but take the cleavage position from the neural networks part "but it has a slightly lower accuracy in predicting the precise location of the cleavage site". The HMM result also has the advantage as well of being less ambigious ie it makes a clear call rather than leaving it to you to decide how many yes's constitutes a positive result. Hope this helps...
Sarah Sarah Larkin McQuaid, PhD. Sr. Scientific Investigator/Bioinformatics Group Leader, Dept. of Cell and Molecular Biology, Amylin Pharmaceuticals, Inc. Phone: 858.642.7357 http://www.amylin.com -----Original Message----- From: Kerrie Dunstan [mailto:[EMAIL PROTECTED]] Sent: Wednesday, December 11, 2002 3:26 PM To: [EMAIL PROTECTED] Subject: [artemis-users] signalP anyone?? You guys are my last resort! Sorry this is not directly artemis-related. Has anyone been using SignalP? If so do you use the HMM of the Neural network model for predicting your signal peptides or both? Have you noticed how the results from each model are almost always conflicting? Even if they both predict a signal peptide they almost always give different cleavage sites, often by only 1 or 2 amino acids. If anyone knows whether one model is more reliable than the other or has also encountered this problem I would appreciate your feedback! I have emailed the Signal P distributors countless times and never get a reply. Thanks! -Kerrie Dunstan http://greetings.yahoo.com.au - Yahoo! Greetings - Send your seasons greetings online this year!
