3dDNA: A Computational Method of Building DNA 3D Structures

Abstract

Considerable progress has been made in the prediction methods of 3D structures of RNAs. In contrast, no such methods are available for DNAs. The determination of 3D structures of the latter is also increasingly needed for understanding their functions and designing new DNA molecules. Since the number of experimental structures of DNA is limited at present, here, we propose a computational and template-based method, 3dDNA, which combines DNA and RNA template libraries to predict DNA 3D structures. It was benchmarked on three test sets with different numbers of chains, and the results show that 3dDNA can predict DNA 3D structures with a mean RMSD of about 2.36 Å for those with one or two chains and fewer than 4 Å with three or more chains.

Keywords: 3D structure prediction; 3D template libraries; 3dDNA; DNA.

Figure 1

Comparison of the prediction accuracies (all-atom RMSDs) of 3dDNA for the assembled, optimized, and best structures of 31 single-chain DNAs with or without open loops.

Figure 2

The structural analysis of open loop in a DNA (PDB ID 1EN1). (A) The secondary structure of 1EN1. (B) The native, assembled model and optimized structures are marked with grey, green, and blue, respectively. The first model in NMR selected as native. When the open loop is not considered, the RMSD of the assembled structure drops from 8.32Å to 4.2Å. (C) Twenty models obtained from the NMR experiment, of which model 1 is marked with grey. The 2D and 3D structures are generated by Forna [37] and PyMOL [38], respectively.

Figure 3

Two example of predicted 3D structures by 3dDNA. (A) The 2D and 3D structures of DNA 4KB1_1 with perfect template; the RMSD of assembled (green) and optimized (blue) structures with a native one (grey) are 0.56 Å and 5.3 Å, respectively. (B) The 2D and predicted 3D structures of DNA 1OMH with imperfect template, the RMSD of assembled (green) and optimized (blue) structures with a native one (white) are 9.8 Å and 6.5 Å, respectively.

Figure 4

Comparison of the prediction accuracies (all-atom RMSDs) of 3dDNA for the assembled, optimized, and best structure of double-chain DNAs, respectively.

Figure 5

Two examples of the predicted 3D structures of double-chain DNA by 3dDNA. (A) The 2D and 3D structures of DNA 3RB6_1 with perfect template, the RMSD of assembled (green) and optimized (blue) structures with a native one (grey) are 0.93 Å and 4.95 Å, respectively. (B) The 2D and 3D structures of DNA 4DAV with imperfect template; the RMSD of assembled (green) and optimized (blue) structures with a native one (grey) are 10.77 Å and 4.92 Å, respectively.

Figure 6

Comparison of the prediction accuracies (all-atom RMSDs) of 3dDNA for assembled, optimized, and best structures of DNAs with multi-chains.

Figure 7

Two examples of the 3D structures of multi-chains DNA predicted by 3dDNA. (A) The 2D and 3D structures of DNA 1TW8 with perfect template; the RMSD of assembled (green) and optimized (blue) structures with a native one (grey) are 0.9Å and 8.0 Å, respectively. (B) The 2D and 3D structure DNAs 4L74 with imperfect template; the RMSD of assembled (green) and optimized (blue) structures with a native one (grey) are 2.8 Å and 6.8 Å, respectively.

Figure 8

Comparison of the prediction accuracies (all-atom RMSDs) of the assembled, optimized, and best structures of 3dDNA with previous indirect predictions by Jeddi.

Figure 9

The proportion of five classes of DNA structures in PDB. The 5 classes include unStruck-DNA (DNA without base pairs), Helix-DNA (DNA with pure duplex structure), D-DNA (DNA with both duplex and loop structures), T-DNA (DNAs containing triple helices), and G-DNA (DNAs containing quadruple helices or consecutive stacking G-quadruples).

Figure 10

The workflow of decomposing DNA 1QSK to five SSEs. Break loop, bulge loop, and open loop are marked by red, grey, and yellow rectangular boxes, respectively. (A) The secondary structure of DNA 1QSK. (B) All the SSEs, which together form a secondary structure tree (SST) of DNA 1QSK, including two helices and one open loop with 2D structure “.(()).”, a bulge loop with 2D structure “((…..(())))”, and a break loop with 2D structure “((.&.))”. DNA 2D plots are generated using Forna [37].

Figure 11

Workflow chart of 3dDNA for DNA 3D structure prediction. Mainly composed of three parts: Templates Searching, Assemble, and Optimization.