In this work, we made use of the target dictionary provided by Nagamine and Sakakibara [18] to represent the biding sites. The dictionary is organized as three layers: the amino acids, trimers and trimer clusters. In the first layer, each amino acid is first represented by 237 items of physical-chemical properties [19,18], such as residue volume, polarizability and solvation free energy. And then, a principal component analysis (PCA [20]) is applied to reduce the dimension. As a result, each amino acid is described by a 5-dimensional feature vector. In the second layer, twenty amino acids are permutated and combined into 4200 trimers. Each trimer αtri(α01, α11, α12) is mapped into a 5-dimensional vector space [18] as follows: