Table 6.
High-affinity prediction | Specificity prediction | ||||||
---|---|---|---|---|---|---|---|
Modelsa | Library SPOT AUC |
Substitution SPOT AUCb |
Modelsa | Library SPOTc AUC (R) |
Yeast screeningd AUC |
||
Bim | Bad/Noxa | ||||||
STATIUMX | 0.94 | 0.72 | 0.66 | STATIUMX and STATIUMM | 0.9 (0.86) | 0.99 | |
STATIUMX_Mstruc | 0.82 | 0.67 | 0.64 | STATIUMX and STATIUMM_Xstruc | 0.81 (0.78) | 0.9 | |
STATIUMM | 0.94 | 0.71 | 0.69 | STATIUMX_Mstruc and STATIUMM | 0.52 (0.21) | 0.37 | |
STATIUMM_Xstruc | 0.95 | 0.65 | 0.71 |
Two STATIUM models were tested: Bcl-xL sequence with an Mcl-1 structure model (STATIUMX_Mstruc) and Mcl-1 sequence with a Bcl-xL structure model (STATIUMM_Xstruc). Specificity prediction (right column) requires two models, because specificity is predicted using a difference in scores.
The AUC values in the left column are for SPOT data from Bcl-xL or Mcl-1 binding Bim BH3 variants, and the AUC values in the right column are for SPOT data from Bcl-xL binding Bad BH3 variants and Mcl-1 binding Noxa BH3 variants.
The values in parentheses are the Pearson correlation coefficients for the same data set.
AUC value for prediction of the specific sequences used in Fig. 4a, when one class is Bcl-xL specific (40 sequences) and the other is Mcl-1 specific (33 sequences) and the difference in energies is used to predict the state.