Feature Synthesis and Extraction for the Construction of Generalized Properties of Amino Acids

Amino acid similarity matrices are used for protein sequence comparison. It has been shown previously that they can be reconstructed from equivalence classes between amino acids. The goal of the current study is to propose an algorithm for identification of the properties that generate these equivalence classes. An approximate reasoning method for feature extraction and synthesis is developed to this end. It is shown that these equivalence classes are related with the amino acid properties that are important for the formation of the protein structure. The algorithm presented in this study works best for bit-patterns, which are frequently encountered in bit-vector representations.