Figure 5

The ten most important (highest “GAIN”) features considering only the basic features (A) and all features (B). The features from top to bottom: sasaCompatibility—a measure of the agreement between the solvent accessible surface area of the model’s residues (as measured by DSSP⁶⁵) and their predicted accessibility⁵⁶. goap_ag—a pairwise orientation-dependent knowledge-based potential⁴⁰. deepCNF8Compatibility—a measure of the agreement between the secondary structure (8 states) of the model’s residues (as measured by DSSP⁶⁵) and their predicted secondary structure⁶⁶. contacts8 and contacts14—the average numbers of contacts with thresholds of 8 Å and 14 Å, respectively, between carbon atoms. meshinr_dssp8Compatibility and meshirw_dssp8Compatibility_Weighted—two slightly different measures of the agreement between the secondary structure (8 states) of the model’s residues (as measured by DSSP⁶⁵) and their predicted secondary structure⁵⁹. scCarbonN—the number of carbon atoms in the model’s side-chains. coverage—the fraction of the target sequence, which the are modeled. SheetFraction—the fraction of beta-sheet resides within the residues with any secondary structure. consensus features—see Eqs. (1–4). gdt1_consensus_median—the median value of gdt1_consensus, among all the models of a specific target. hydrogenBondsPairs_median—the median value of a cooperative hydrogen bonds energy term⁶⁷ among all the models of a specific target. cooperativeZstdRamachandranSidechain_median—the median value of a cooperative torsion angle energy term, among all the models of a specific target.