Abstract

The attachment of glycosylphosphatidylinositol (GPI) is one of the most important post-translational modifications of proteins and plays an important role in promoting biochemical activities in eukaryotic cells. GPIanchored proteins (GPI-APs) are characterized by GPI-anchor attachment signals of hydrophobic residues and small residues near the GPI-anchoring site (ω-site). Here, we describe a new method for predicting GPI-APs based on hydropathy profiles and position-specific scores (PSSs) in combination with the back propagation artificial neural network (BP-ANN). First, the sequences of GPI-APs and negative controls were aligned according to residue size in the Cterminal region and the position-specific amino acid propensities were analyzed according to their alignment positions. Next, PSSs were created using the amino acid propensities of GPI-APs and the negative controls, and BP-ANN with a three-layered structure was trained by the PSSs. The accuracy of discriminating GPI-APs from the negative controls was evaluated in a 4-fold cross-validation test and GPI-APs were detected with 92.9% sensitivity and 94.8% specificity. This result shows that our method can predict GPI-APs with high accuracy and a combination of PSSs and BP-ANN can effectively discriminate GPI-APs.

Keywords: Back-propagation artificial neural network (BP-ANN), discrimination, GPI-anchored protein (GPI-AP), GPI attachment signal, position-specific scoring matrix (PSSM), post-translational modification, position-specific scores, C-terminal amino acids, FT LIPID, N-terminus