Accuracy of clustering in Training and Test-18 sets.

†) TP: FLIP found in Cluster 1TN: FUNC found in Cluster 2
FP: FUNC found in Cluster 1FN: FLIP found in Cluster 2
The accuracy and Matthews correlation coefficient (MCC, a measure of the quality of a binary classification) of the results of the clusterings shown in Figure 4 are indicated. The overall accuracy is 76% and 78% for both training Test-18 sets, respectively. TPs are quite readily identified in both training and Test-18 sets (80% and 69% sensitivity, respectively). The majority of TPs are enzymes and immunoglobin heavy chain-light chain interactions. TNs are less well identified (70% and 56% negative predictive values, respectively). MCCs of 0.50 and 0.62 indicate that our simple two-category approach is generally appropriate.

†) 真阳性（True Positive, TP）：FLIP 见于簇1；真阴性（True Negative, TN）：FUNC 见于簇2；假阳性（False Positive, FP）：FUNC 见于簇1；假阴性（False Negative, FN）：FLIP 见于簇2。
图4展示的聚类结果的准确率与马修斯相关系数（Matthews Correlation Coefficient, MCC，一种衡量二分类任务质量的指标）已在文中标注。训练集与Test-18测试集的总体准确率分别为76%与78%。真阳性样本在两类数据集上均较易识别，灵敏度分别达80%与69%，其中多数真阳性样本为酶与免疫球蛋白重链-轻链相互作用对象。真阴性样本的识别效果相对欠佳，阴性预测值分别为70%与56%。两组MCC值分别为0.50与0.62，表明我们所采用的简单二分类方法整体适用性良好。