Figure 9

Three examples of apparent bias of EMA prediction, by the absence of molecular context. The figure presents the large (2225 residues) DNA-dependent RNA polymerase of crAss-like phage phi14:2 (PDB 6VR4) that gave rise to ten CASP14 targets: The whole protein (T1044), which was not offered as an EMA target, and nine domain targets T1031, T1033, T1035, T1037, T1039, T1040, T1041, T1042, and T1043. For six of them: T1031, T1033, T1035, T1039, T1040, and T1043, MESHI_consensus failed to provide reliable EMA predictions (LOSS above 0.16). We speculate that these failures may be attributed, at least partially, to the absence of the protein context in the isolated domains. (a) The three domain targets, depicted in the context of the whole protein (yellow): T1031 (residues 1–95, red), T1033 (residues 96–196, cyan), T1040 (residues 1372–1501, magenta). (b–d) The three interfaces of these domains (respectively) with the rest of the protein. A sample of the interacting residues is shown, the rest are hidden for clarity. Blue dashed lines represent hydrogen bonds and salt bridges (Figure is drawn with ChimeraX⁶⁸).