Shown here are three confusion matrices, thetop (a) tested using the full feature vector description (listed in Table1), the middle (b) using only the most local features, and the bottom (c) using only geometric features. Each row represents the tests for a ligand classifier run on all test cases. Each column represents an individual testing example, grouped by the ligand the protein is known to bind to. The value in the cell is the area underneath the precision/recall curve produced from that test. A higher value indicates a better match. Green cells indicate true positive results: the predictor found the ligand it was trained for. Purple cells indicate false negatives: the ligand failed to find the ligand it was trained for. Red cells indicate false positives: the predictor found the site of a different ligand. See Figure 2 for illustrations to help interpret these numbers.
Cipriano et al. BMC Bioinformatics 2012 13:314 doi:10.1186/1471-2105-13-314