[Protein structure prediction from analogy. III. Optimization of a combination of substitution matrices and pseudo-potentials used to align protein].

We describe a general and fast method of maximization of a "recognition ability" of a linear combination of an arbitrary number of various methods used to recognize protein structures and produce the sequence-to-structure alignments. It is shown that, at a low level of sequence similarity, the optimal combined method has a significantly higher "recognition ability" than each of participating separate method, the leading role in this combination being played by (1) pseudo-potentials of long-range interactions; (2) matrices of secondary structure similarity; and, finally, (3) amino acid substitution matrices. As to a high level of sequence similarity, here the leading and virtually the sole role in the optimal combination is played by substitution matrices, although admixture of pseudo-potentials of long-range interactions and matrices of secondary structure similarity somewhat increases the "recognition ability" of the combined method.