Transmembrane protein sequences are badly conserved during evolution.Even two homologous proteins have a low level of sequence identity.Consequently,the commonly used method to select training sequences based on sequence identity can not efficiently reduce the sampling bias in the transmembrane segment predictions.To solve this problem,this paper presents a new prediction algorithm based on fuzzy cluster analysis of amino acids.It clusters the amino acids into groups according to their distribution similarity in different regions and then makes the prediction based on the distribution properties of each group instead of those of each amino acid.The results show that the new algorithm can efficiently reduce the impact of the selection of training sequences on the prediction results to some extent and thus improve the prediction accuracy.

Key words: amino acid, transmembrane protein, fuzzy cluster, transmembrane segment prediction