Figure 1

Considering background distributions for individual subjects improves detection quality. The standard approach based on the query calibration only (black ROC curve) is compared to the schemes that involve combining the query-based score distribution with individual subject-based distributions, produced by the comparison to the calibration database (green), to the set of all non-homologs in the search database (red), or, as controls, to the sets of randomized database profiles with shuffled positions: blue, profiles with the secondary structure assignment the same as in the real profiles; cyan, profiles with artificial secondary structure, 'coil' assigned to all positions. Combination of distributions produced by the comparison of query to the calibration database and of subject to the full set of its non-homologs results in the highest performance (red curve).