An Integrative Multi-Network and Multi-Classifier Approach to Predict Genetic Interactions (original) (raw)
Figure 4
Receiver operating characteristic (ROC) curves of our ensemble based on SL-dependent (MNMC.all) and –independent features (MNMC.slif).
(A) ROC curves (AUC of MNMC.all = 0.819 and MNMC.slif = 0.851) and (B) Precision-recall curves for classification of the 337 least connected SL and 199 corresponding non-SL interactions using our SL-independent and SL-dependent features; (C) ROC curves (AUC of MNDT.all = 0.862, MNMC.all = 0.897, MNDT.slif = 0.598 and MNMC.slif = 0.805) and (D) Precision-recall curves for classification on Wong et al.'s SSL dataset [8] using MNMC and MNDT based on on either all the features (all) or the SL-independent features (slif); (E) ROC curves (AUC of MNMC.all = 0.616 and MNMC.slif = 0.633) and (F) Precision-recall curves for classification on an independent test set constructed from the SGD interaction database using MNMC.all and MNMC.slif. The corresponding ROC and precision-recall curves for a random classifier applied to all these prediction problems are also shown.