Data set | Unstable | Non-assigned | Stable |
|---|
Training/testing data | 795 | 2442 | 743 |
Predictions 1 (P1) | 6974 | 10683 | 6357 |
Predictions 2 (P2) | 3319 | 17475 | 3220 |
- The first row shows the number of proteins in each of the stable, unstable and non-assigned groups that were allotted based on the cluster analysis of the GPSP data [1]. This data was employed in all of the statistical analyses and the proteins in the stable and unstable classes were used for training and testing the models. The second and third rows show how the predictions made by the BN+SVM model trained on the full data set (P1) and the predictions made by the BN+SVM model trained on the trimmed data set (P2) can be assigned to stability classes when using thresholds to define predicted “stable” and “unstable” proteins. For P1, a stable protein was defined as scoring above 0.75 and an unstable protein had to score below 0.2. For P2, a stable protein was required to score above 0.7 and an unstable protein to score below 0.3. For both P1 and P2, if a protein scored above the “unstable” threshold and below the “stable” threshold it was classified as “non-assigned”. These thresholds were set such that the number of proteins predicted to fall into the three different stability groups reflected the numbers from the cluster analysis on the GPSP data.
