How much are homologous peptides homologous?

A statistical analysis designed to deal with the problem of identifying homologous pairs of "short sequences of amino-acids" (= peptides) belonging to different proteins is presented. The conceptual novelty of the searching strategy proposed here lies in the fact that both the degree of homology of the two peptides of the pair (measured by a suitably defined affinity score) and the level of statistical significance of its occurrence are taken into account on equal footing. They are combined in a sort of "biological indicator", characterising each pair. Pairs for which the value of the biological indicator is larger than an appropriate threshold are taken as statistically significant and (putatively) biologically relevant. The method is employed in various test cases and proves to be reliable and efficient. In particular we have studied the cases in which the known existence of an auto-immune response has lead to the identification of homologous peptide pairs between human and viral or bacterial proteins. The detection efficiency of the algorithm in these cases turns out to be especially good when the most naïve affinity table, the Identity matrix, is employed to measure the similarity of amino acidic pairs. In contrast, when the 250-PAM mutation matrix is used, the detection efficiency goes to zero.