Comparison of methods.

Each cell contains performance, sensitivity, specificity, and the number of unknown domains flagged correctly (in percent) in that order on the CATH v3.2.0 test set at each CATH level for all three methods. Decision thresholds were calibrated on the CATH 3.2.0 training set to achieve 95% sensitivity. Levels comprising only a single domain in CATH 3.2.0 are never included in the training set and thus account for the numbers in the unknown column; for C- and A-levels, there are none of these.

在CATH v3.2.0测试集的各CATH层级下，三种方法的评估结果按顺序填入每个单元格，依次为整体性能、灵敏度（Sensitivity）、特异度（Specificity），以及正确识别的未知结构域占比。所有决策阈值均在CATH 3.2.0训练集上校准，以达到95%的灵敏度要求。在CATH 3.2.0中仅包含单个结构域的层级从未被纳入训练集，此类层级的占比即为未知列中的数值；而C层级与A层级则不存在此类单结构域的情况。