- Many proteins contain a signal sequence at their COOH-terminus recognized by glycosylphosphatidylinositol (GPI) anchor and attached on the membrane. Experimental result suggests that the overall hydrophobicity of COOH-terminus is more important than precise sequence. We use a machine learning technique, support vector machine, to examine requirement for identifying this signal sequence measured at hydrophobicity scale by computational methodology. Effects of hydrophobicity of the signal sequence at multilevel and different segments of the signal motif proposed by previous works on their identification were investigated. For precisely identifying the signal sequence, we found that 15 residues (a hydrophobic domain) proximity to the COOH-terminus of the signal sequence is necessary. This result is consistent with observations from experimental methodology. Moreover, aside from 15 residues, the segment of 40 residues on COOH-terminus also contributes to its precise identification.