Functional relevance of CASP16 nucleic acid predictions as evaluated by structure providers

Key results

1. 01

   ShownPredictions recovered helices and much of the secondary structure but relied on canonical A-form geometry, missing sharp backbone bends and non-canonical pairs. For the xrRNA, most high-scoring models passed the 3′ end through a closed ring, forming an implausible knot.

   xrRNA R1205 (2.9 Å crystal): lowest RMSD 7.46 Å (RNApolis TS063_4) but a 6-bp instead of 9-bp pseudoknot; Vfold TS481_3, best by most CASP metrics (GDT-TS 0.43, LDDT 0.499), formed a knot, as did AlphaFold3, GromihaLab, KiharaLab and GuangzhouRNA models; no model reproduced the cis A-minor contacts. · 2.1 Exoribonuclease-resistant viral RNA (CASP…; 3 Conclusions

2. 02

   ShownFunctionally decisive contacts were missed even when the global fold was right. Most group IIC intron models with an accurate fold lacked the D2-D6 tertiary contacts, and no SAMURI ribozyme model reproduced the active site or the SAH binding mode.

   Group IIC intron R1241: of 180 models, 105 had RMSD <2 Å, yet 101 of these missed the η-η’ and π-π’ contacts; only 4 (three Vfold, one GeneSilico) recovered them. SAMURI R1288: LDDT 0-0.16 for all models; no model exceeded 4 out of 10 manual points; molecular replacement failed (best translation function Z-score 7). · 2.5 SAM utilizing alkyltransferase ribozyme…; 2.7 Oceanobacillus iheyensis Group IIC intron…; 3 Conclusions

3. 03

   ShownFor HIV-1 RRE stem-loop II, individual subdomains were modeled closely but their relative orientation at the three-way junction was poor for the wild type and better for the G34U mutant, whose simpler junction the providers suggest eased prediction.

   Top-five global RMSDs were 7.8 to 9.4 Å for SLII R1209 (213 models) vs 3.08 to 8.23 Å for SLIIG34U R1296 (230 models); Vfold R1296TS481_3 reached 3.08 Å. Subdomain RMSDs for the R1209 top five: IIa 0.856 ± 0.213 Å, IIb 2.899 ± 0.284 Å, IIc 0.786 ± 0.230 Å. · 2.2 HIV-1 rev response element stem-loop II…; 3 Conclusions

4. 04

   ShownLigand-dependent conformations and ligand contacts were poorly modeled. Riboswitch ligand pockets with a template were acceptable, but models missed a nucleotide shift caused by a bulkier ligand and the unusual fold of a dopamine-bound DNA aptamer.

   ZTP riboswitch: 72% and 77% of models had TM-score >0.45 for R1261/R1262, but only 4 (1.8%) and 1 (0.5%) exceeded 0.70; no group exceeded local lDDT 0.70 for C52 in R1264. DNA aptamer D1273 (27 nucleotides): 107 models from 22 groups, RMSD 10.16 to 27.79 Å; dopamine stacking was partly recovered, hydrogen bonds were not. · 2.4 ZTP riboswitches bound to synthetic and…; 2.6 Dopamine-bound DNA aptamer RKEC1 (CASP…; 3 Conclusions

5. 05

   ShownIntermolecular interfaces were mostly mispredicted. Across RNA dimers, an aptamer-protein complex and protein-DNA complexes, models often placed the partner in the wrong region or missed the specific contacts; none of the leading CbUDE models captured the arginine intercalating into ssDNA.

   OLE dimer R1285o: at most one of three interfaces per model; 20 models got the kissing-loop B3, 6 a B2-like contact, none the B1 A-A pairs. 2′ FY aptamer M1282: 36 % bound the correct top of the RBD, 56 % the back. SPβ M1239: 35 of 41 models had the tetrameric core, but none matched the CC-motif angle. · 2.9 Ornate Large Extremophilic (OLE) RNA Dimer…; 2’ FY RNA aptamer / protein complex (M1282); 2.11 Cryo-EM structure of SPβ LSI-RDF synaptic…; 2.12 Cryptic DNA-binding protein UDE from…