Optimal dictionaries of the final information on the basis of the solvability criterion and their applications in bioinformatics

Abstract

Within the algebraic approach to the pattern recognition, the sets of the input (initial) and the output (final) information are defined on the basis of some additional expert information from the relevant subject area. Finding, in a sense, “optimal” definitions for the sets of the output information is crucial, for instance, for solving problems in bioinformatics. In this paper, we propose a formalism to establish optimal dictionaries for output information on the basis of the solvability criterion for the relevant problem. The formalism developed is applied to the problem of recognition of the secondary structure of proteins. The experiments have shown that the optimal dictionary to describe the secondary structure of proteins consists of four five-letter elements. The dictionary allows making a significant increase in the cross-validated recognition accuracy of the secondary structure.