Figures

Figure 1:

GDT_TS Score of all targets predicted by POEM in comparison to the best prediction. Our models in FM category are comparable to best predictions. Also shown are the best models for each of target in our decoy set. Our models for targets T0300, T0335, T0304, T0307, T0309 are among top ten models. Except for T0335 all other targets are in FM category(See also table 1).

Figure 2:

Models for FM targets T0300 (5.51 Å), T0304 (8.14 Å), T0348 (6.52 Å) and T0350 (5.37 Å) are shown. The native structures are shown in green and our predictions in magenta.

Figure 3:

Models for TBM targets T0283 (5.76 Å), T0327 (5.11 Å), T0335 (1.61 Å) and T0354 (6.81 Å) are shown in aligned overlay with the corresponding experimental structures. The native structures are shown in green and our best predictions in magenta.

Figure 4:

Models for HA-TBM targets T0311 (2.74 Å) and T0367 (4.66 Å) are shown with corresponding experimental structures. The native structures are shown in green and our best predictions in magenta.

Figure 5: (missing)

Correlation plot for GDT_TS scores of all server models and those selected by POEM-QA models. GDT_TS scores for server predictions are shown as magenta dots, best ranked structures (BRS) as blue squares and best energy structures (BES) as green triangles. For most of the targets, BRS models corresponds to the best server models, while the BES models are rank high in the population but fail to identify the very best prediction.

Figure 6:

GDT_TS comparison of POEM-QA, POEM-REFINE and best server predictions. Predictions for POEM-QA contains best of BRS/BES models. Notice that POEM-QA identifies the best server prediction in most of the cases. POEM-REFINE predictions are for 24 targets while POEM-QA predictions are for 48 targets.

Figure 7:

Illustration of the “added-value” due to our physics based approach. The left panel shows the GDT_TS scores(shown in red) of the best model using the standard clustering scheme of Rosetta and corresponding POEM-REFINE predictions(shown in blue) for 25 Prediction targets. The right panel shows the comparison between best model selected using a hierarchical clustering procedure for Server models(shown in red) and corresponding POEM-QA predictions(shown in blue) for 48 QA targets.