RNA-Puzzles: a CASP-like evaluation of RNA three-dimensional structure prediction

Abstract

We report the results of a first, collective, blind experiment in RNA three-dimensional (3D) structure prediction, encompassing three prediction puzzles. The goals are to assess the leading edge of RNA structure prediction techniques; compare existing methods and tools; and evaluate their relative strengths, weaknesses, and limitations in terms of sequence length and structural complexity. The results should give potential users insight into the suitability of available methods for different applications and facilitate efforts in the RNA structure prediction community in ongoing efforts to improve prediction tools. We also report the creation of an automated evaluation pipeline to facilitate the analysis of future RNA structure prediction exercises.

FIGURE 1.

Problem 1—An RNA dimer. (A) Secondary structure of the reference RNA molecule. Note that the structure is symmetric on the sequence level but is asymmetric in the crystal, indicating that crystal-packing forces played a significant role in the conformation of this RNA. The interaction (in magenta) was detected with the RNAView (Yang et al. 2003) annotation program but not with MC-Annotated (Gendron et al. 2001). (Thick gray band) Coaxial stacking between helices. X-ray structures of the reference RNA molecule (green) and the lowest RMSD predicted models (blue) for the complete Das model 3 (B), details of loop L1 of Das model 1 (C), and details of loop L2 of Das model 3 (D). (E) Deformation Profile values for each of the five domains of the homodimer. (Colored lines) The DP values for the two predicted models with lowest RMSD, Das model 3 (dark red), and Das model 1 (dark green), and for the predicted model with highest RMSD, Dokholyan model 1 (dark blue). (Radial red lines) The minimum, maximum, and mean DP values for each domain.

FIGURE 2.

Problem 2—An RNA square. (A) Secondary structure of the reference RNA molecule. X-ray structures of the reference RNA molecules (green) and the predicted models with lowest RMSDs (blue) for the full molecule and Bujnicki model 2 (B); details of helices H1, H2, and H4 of Das model 1 and helix H3 of Bujnicki model 2 (C); and details of loops L1 and L2 of Santalucia model 1, loop L3 of Dokholyan model 1, and loop L4 of Bujnicki model 3 (D). (E) Deformation Profile values for the three predicted models with lowest RMSD: Bujnicki model 2 (dark red), Bujnicki model 3 (dark green), and Das model 1 (dark blue). (Radial red lines) The minimum, maximum, and mean DP values for each domain.

FIGURE 3.

Problem 3—A riboswitch domain. (A) Secondary structure of the reference RNA molecule. (B) X-ray structure of the reference RNA molecule (P1 [red]; P2 [orange]; P3, P3a, P3b [yellow]; active site [green]) and the predicted model of the lowest RMSD Chen model 1 (blue). (C) Detail of the ligand binding pocket for the X-ray structure (green) and the lowest RMSD Chen model 1 (blue). (D) Deformation Profile values of the pairwise helical interdomain regions (P1, P2, P3, P3a, and P3b) for the three models with lowest RMSDs: Chen model 1 (dark red), Dokholyan model 2 (dark green), and Das model 5 (dark blue). (Radial red lines) The minimum, maximum, and mean DP values for each interdomain pair.