CHDOCK: a hierarchical docking approach for(3)

Fig.3 The success rate as a function of the number of top predictions for our CHDOCK and CHDOCK_lite tested on our protein docking benchmark of 212 Cnsymmetric complexes for bound docking (A) and unbound docking (B)

Examples of the docking model

Figure 4 shows the top binding modes predicted by our CHDOCK for both bound and unbound docking on three example can be seen from the figure that the predicted complexes overlap well with the experimental native structures,and give a ligand RMSD of 0.42 and 4.03 ? for C2symmetric target 1MSC,0.92 and 3.38 ? for C4symmetric target 1OVO,and 0.95 and 1.20 ? for C6symmetric target 1KQ1, good consistency between the predicted and native structures in both bound and unbound docking demonstrates the reliability of our CHDOCK.

CONCLUSION

We have developed a hierarchical docking algorithm for predicting the complex structures of homo-oligomers with Cnsymmetry,which referred to as Cnsymmetric binding modes were first generated by an FFT-based docking algorithm,in which a shape complementarity scoring function was used to consider long-range ,the binding modes with best shape complementarity were optimized with our iterative scoring function for protein-protein symmetric docking algorithm CHDOCK was evaluated on a diverse benchmark of 212 Cnsymmetric protein complexes from the PDB,and was compared with three state-of-the-art symmetric docking approaches including M-ZDOCK,SAM,and shows that CHDOCK achieved a significantly better performance than the other three docking methods in both the number and the quality of successful predictions for bound docking and unbound results demonstrate the strong predictive power of our hierarchical docking algorithm CHDOCK in modeling Cnsymmetric protein complexes.

MATERIALS AND METHODS

FFT-based translational search

Fig.4 Comparisons between the top predicted binding modes and native structures for three targets 1MSC (C2 symmetry) (A),1OVO(C4 symmetry)(B)and 1KQ1(C6 symmetry)(C).The native structure is colored in pink and the predicted structure is colored by each column,the upper and lower ones are for bound docking and unbound docking,respectively

The putative symmetrical complexes were constructed from a monomer or subunit in 3D translational space by a modified version of our general FFT-based docking algorithm (Yan et al.2017; Yan and Huang 2018).Specifically,we first made two copies of the subunit or was called ‘receptor’ subunit and the other ‘ligand’’ docking with Cnsymmetry,the receptor subunit was fixed and the ligand subunit was rotated by an angle of 360°/n around the perform an FFT-based search,both the receptor and ligand subunits needed to be mapped onto a 3D grid of N×N×N grid points (Chen and Weng 2003;Katchalski-Katzir et al.1992).The grid points within the VDW radius of any protein atoms were considered inside the molecule,and the others were considered as outside the ,the VDW radii for standard protein atoms were taken from the study by Li and Nussinov (1998).Then,the inside-protein grid points were divided into three parts∶ surface layer,nearsurface layer,and core is defined that a grid point belonged to the surface layer if any of its neighboring grid points is outside the ,a grid point belonged to the near-surface layer if any of its neighbors is in the surface the other grid points except the surface and near-surface layers inside the protein were defined as the core to the above definitions,one can see that the nearsurface layer and core region were normally occupied by the protein atoms,and the surface layer separated the inside protein from the outside ,each grid point for the receptor (R) and ligand (L) subunits was assigned a complex value as∶

and

where J2=-1,l,m,and n are the indices of the 3D grid(l，m，n=1，???，N),and r is the distance between the grid points of(i,j,k)and(l,m,n).Here,i ∈[l -3，l+3],j ∈[m-3，m+3] and k ∈[n-3，n+3] for the surface layer,and i ∈[l -1，l+1],j ∈[m-1，m+1] and k ∈[n-1，n+1] for the near-surface layer, also,the grid point (i,j,k) should belong to nearsurface layer or protein core.

With the above mapping of the proteins on the grid,the shape complementarity score between two neighboring subunits of a symmetric complex around the zaxis can be generally expressed by the following equation (Chen and Weng 2003; Katchalski-Katzir et al.1992)∶

where o and p are the number of grid points by which the ligand (L) is shifted with respect to the receptor(R)in the x-y plane,re is no shift in the z-axis because the rotational axis is parallel to the z-axis,which reduces the sampling space in one translational correlation of Eq.3 can be calculated by an FFT-based higher correlation score means a better shape complementarity between two grids for a relative translation of (o,p) (Katchalski-Katzir et al.1992).

Rotational sampling strategy

To perform a global sampling approach for putative binding modes,one needs to search the six-dimensional(i.e.,3 translational + 3 rotational) exhaustive search in 3D translational space can be performed by an FFT-based approach,as described in the previous exhaustive search in the rotational space will be conducted in the space of Euler angles by taking into the Cnsymmetry restriction ,the monomer subunit is rotated by an interval of Euler angles (φ=0,Δθ,Δψ) in the rotational space,where the angular definition is based on the so-called ‘x-axis convention’.Namely,φ is the first rotation about the zaxis,θ ∈0，π/2[ ]is the second rotation about the former x-axis (now x′),and ψ ∈(0,2π] is the third rotation about the former z-axis (now z′).It is unnecessary to sample the φ angles as the rotational axis is addition,θ only needs to be sampled within 0，π/2[ ]instead of 0，π[ ] because of the rotational these reduce the sampling space in the rotational space.

文章来源：《科学学研究》 网址: http://www.kxxyj.cn/qikandaodu/2020/0925/391.html