Figure 4

Statistics based on the knowledge of all query's non-homologs: a possible hypothetical performance. The detection quality that could potentially be achieved by using the distribution of query's scores to all non-homologs is compared to the performance of other schemes. ROC curve for the query-based distribution produced on the full set of non-homologs (cyan) is similar to that for the subject-based distribution produced on all non-homologs alone (blue) or mixed with query calibration distribution (red). This performance can be additionally improved by combining both query- and subject-based distributions generated on the full sets of their non-homologs (green). Black, ROC curve for the query calibration alone.