Figure 2.

Performance of vScreenML on a test dataset. (A) Discriminating the performance of the original vScreenML model (left) versus the new vScreenML 2.0 (right). The vScreenML score for each data point was calculated using cross-validation, where each fold was prepared by clustering protein sequences using MMSeq2. Active complexes (true class label is 1) are represented by blue data points and decoy complexes (true class label is 0) are represented by red data points. The orange dashed line indicates an equal likelihood of being classified as active or decoy. (B) Receiver operator characteristic plots showing performance of vScreenML 2.0 relative to the original model.