“…For instance, the residues at positions -3 and -1 relative to the cleavage site are usually small and neutral. The regular expression search and weight matrix algorithms are now replaced by more sophisticated methods that include various types of machine learning methods such as neural networks (Nielsen et al, 1997), support vector machines (Vert, 2002), hidden Markov models (Nielsen et al, 1999) and many others (Ladunga et al, 1991;Talmud et al, 1996;Nielsen et al, 1997a;1997b;Nielsen & Krogh, 1998;Nielsen et al, 1999;Bendtsen et al, 2004a;Menne et al, 2000;Chou 2001;Lao et al, 2002a;2002b;Vert, 2002;Juncker et al, 2003;Hiller et al, 2004;Kall et al, 2004;Zhang & Henzel, 2004;Liu et al, 2005;Sidhu & Yang, 2006). However, most of these methods classify proteins as secretory or non-secretory but do not provide cleavage site assignment.…”