Figure 7. Dissecting Prediction Quality. A).

The performance of the top groups (averaged GDT_TS, left panel) is dictated by two components of the provided contact information: the percentage of correct contacts (those within 8 Å in the target structure) over all given contacts (CCP, middle panel) and the fold coverage of correct contacts over the target structure (CCC, right panel). The bars represent the averages over targets from each contact assisted category: predicted Tp (blue), correct Tc (red), simulated NMR Ts (green), and crosslinked Tx (orange). B) The contacts provided for the simulated NMR Ts category can be subdivided into several classes of given information: all provided contacts that include both single peaks and multiple peaks for certain atom pairs (cyan), unambiguous contacts that correspond to given atom pairs with a single peaks (purple), ‘Ts-specific’ true contacts defined as pairs with atomic distance in the target structure within the given upper distance limit (UPL) (green), and correct contacts defined as pairs within 8 Å in the target structure (medium green). Contacts of each subcategory are shown in logarithmic scales and counted by atom (dark colors, labeled ‘Atom’) or by residue (light colors, labeled ‘Res’). C) The various classes of simulated NMR Ts information lead to different levels of performance measured for “dummy” models generated by us using standard NMR structure determination techniques (see Matherials and Methods for details). The GDT_TS scores of these dummy models produced with all contacts, unambiguous contacts, and ‘Ts-specific’ true contacts are colored (from bottom to top) cyan, purple, and green, respectively, and shown as solid lines to aid visualization. Dummy model performance (colored lines) is compared to prediction model performance (GDT_TS) for all groups (blue circles), with the top simulated NMR Ts prediction models (solid red line) outperforming the top dummy models for all targets.