Computational protocol: γ-TEMPy: Simultaneous Fitting of Components in 3D-EM Maps of Their Assembly Using a Genetic Algorithm

Similar protocols

Protocol publication

[…] The method was tested on both simulated and experimental “target” maps of protein assemblies. The simulated benchmark contains a total of ten assemblies (). For each assembly, the method was tested using three different simulated maps at 10, 15, and 20 Å resolution. These maps were produced by blurring the atomic positions of the assemblies using a Gaussian point-spread function with sigma factor of 0.356 (). The voxel sizes of simulated maps were kept to 3.5 Å. The number of components in the assembly ranges from three to eight and the component size ranges between 88 and 525 residues. The experimental benchmark contains four assembly maps taken from the Electron Microscopy Databank (EMDB) (). The EMDB entries for the assembly maps are 1340, 1980, 2355, and 1046 at 9.0, 7.2, 16.0, and 23.5 Å resolution, respectively (). The PDB entries for the fits that correspond to EMDB maps are PDB: 2P4N, 4A6J, 4BIJ, and 1GRU, respectively. For measuring the prediction accuracy we consider a deposited fit as the reference fit (“native fit”). The number of components ranges from three to seven. Below, and in and , we describe the results of running γ-TEMPy for each of the test cases. For illustration purposes, we show in and the results of five examples from the simulated benchmark (PDB: 1CS4, 2B09, 1MDA, 1TYQ, 2GC7, which represent different numbers of components) and all the test cases from the experimental benchmark, respectively. [...] For the best-predicted (BP) assemblies using target maps simulated at 10 Å resolution, the topology score (TS, see for details) ranged between 0.8 and 1.0 (prior to refinement, A). The translation and rotation components of the assembly placement score (APS) () (see for details) ranged from 1.3 to 7.9 Å and 13.3° to 79.9°, respectively (A). The Cα root-mean-square deviation (RMSD) (see for details) between the components of the BP assemblies and the corresponding native assemblies ranged from 3.2 to 16.9 Å (A). In eight of the ten cases (all except PDB: 1MDA and 1SGF), the BP assemblies identified by the GA had correct topology with TS = 1.0. In the case of PDB: 1MDA, only for chain M, the configuration deviated considerably with respect to the native assembly. The translation and rotation values of component placement score (CPS) were 19.6 Å and 82.2°, respectively (A). Similarly for the case of PDB: 1SGF, chain Y deviated considerably with respect to the native (CPS: translation = 27.8 Å and rotation = 81.4°) (A). In 50% of cases (PDB: 1CS4, 2DQJ, 2BO9, 1GPQ, and 2BBK), the topology of the highest-scoring (HS) assembly was correctly predicted, with a TS = 1.0, and in the case of PDB: 1CS4 it was also the BP assembly (A). The BP assembly was found within the top five ranks in eight of ten cases, and in seven of ten cases was found within top three ranks. In all the cases at 10 Å resolution, the fitness values of the native assemblies were always better than the predicted assemblies.Following Flex-EM refinement () (see for details), the APS for the BP assemblies ranged from 0.5 to 8.2 Å for the translational and 1.0° to 77.3° for the rotational score components, respectively (A and A). The refinement helped to reduce the RMSD of the BP assembly of PDB: 1CS4 from 4.0 to 2.8 Å; of PDB: 2DQJ from 3.5 to 0.7 Å; of PDB: 1VCB from 7.7 to 4.8 Å; and of 1GPQ from 3.2 to 0.6 Å (A). In all other cases (four of which had TS = 1 prior to refinement), the refinement resulted in a marginal decrease or increase in RMSD. This is due to the fact that in those cases the starting fits (BP) before refinement deviate (at least in one of the components) considerably from the native structure with a minimum and maximum RMSD of 10.9 and 16.9 Å, respectively. It is worth noting that for the case of PDB: 1MDA (a six-component assembly), the TS improved from 0.8 to 1.0 after Flex-EM refinement (A). Despite this improvement, the RMSD indicated that the model is far from its native configuration (reduced from 14.1 to 12.0 Å). The Cα RMSD with respect to the native component chain IDs J, H, and L was 4.8, 7.3, and 2.9 Å, respectively, whereas the Cα RMSD of the components with chain IDs M, B, and A was 16.5, 22.9, and 17.6 Å, respectively. The CPS (the translation and rotation pair) for the latter three chains was (9.7 Å, 86.9°), (8.1 Å, 155.6°), and (2.0 Å, 157.1°), respectively. Even though all the components were placed correctly, as evidenced by the good TS, the higher RMSD for chains M, B, and A resulted mainly from a rotation of these chains relative to their corresponding native position (B, chains M, B, and A shown in green, red, and yellow, respectively). […]

Pipeline specifications

Software tools TEMPy, Flex-em
Application cryo-EM