Skip to main content

Table 2 Average Accuracy of the Optimal RNA Structure Predicted with Mfold 3.1

From: Evaluation of the suitability of free-energy minimization using nearest-neighbor energy parameters for RNA secondary structure prediction

  5S rRNA 16S rRNA 23S rRNA tRNA
  M1 C2 P13 M C P24 M C M5 C
Sequences 309 90 56 22 496 72 5 256 484 569
Accuracy6,7,8,9 78 ± 23 71 ± 24 46 ± 17 51 ± 16 41 ± 13
45 ± 16
44 ± 11 57 ± 14 41 ± 13
43 ± 12
83 ± 22 69 ± 24
High/Low10   98/0 81/10   77/5 74/19   74/1   100/0
Median   81    41    41   70
Distributions           
≤ 20% acc11   4 9   4 1   6   2
≥ 60% acc12   77 25   9 6   5   60
20%<acc<60%13   19 66   86 93   89   39
  1. All values are percentages unless otherwise indicated. All averages are per sequence averages for folding complete sequences as defined in the Per Sequence Averages section in Methods. C, Current Study; P1, Previous Study by Gutell Lab for 16S rRNA[29]; P2, Previous Study by Gutell Lab for 23S rRNA[30]; M, Previous Study by Mathews et al.[31]. Accuracies from all previous studies are for folding complete sequences.
  2. 1 All sequences from the Mathews et al. study (M) were folded with Mfold 3.1 using a window size (W) of 0, percent suboptimality (P) of 20%, maximum number of suboptimals (MAX) of 750 and efn2 re-evaluation and re-ordering.
  3. 2 All sequences in the current study (C) were folded with Mfold 3.1 using a window size (W) of 1, percent suboptimality (P) of 5% and efn2 re-evaluation and re-ordering
  4. 3 All sequences in the previous Gutell Lab study on 16S rRNA (P1) were folded with Mfold 2.3 using a window size (W) of 10 and no efn2 re-evaluation and re-ordering.
  5. 4 All sequences in the previous Gutell Lab study on 23S rRNA (P2) were folded with Mfold 2.3 using a window size (W) of 20 and no efn2 re-evaluation and re-ordering.
  6. 5 Bases modified in tRNA that are subsequently unable to fit into an A form helix were constrained to be single-stranded.
  7. 6 Comparative base-pairs that are pseudoknotted were excluded from the analysis in previous Gutell Lab studies (P1, P2), but were included in the current study. The Mathews et al. study included a measure of the percentage of pseudoknotted base-pairs in comparatively predicted structures they considered, but it was unclear if they were included in the analysis.
  8. 7 In all studies, only canonical, comparative base-pairs (excluding any base-pairs with IUPAC symbols) were considered. For both the current study (C) and previous Gutell Lab studies (P1, P2), a predicted base-pair was considered correct only if it matched a comparative base-pair exactly. In the Mathews et al. (M) study, a base-pair was considered if: 1. it matched a comparatively predicted base-pair exactly or 2. either nucleotide of the Mfold predicted base-pair (X,Y where X and Y are the positions of the nucleotides in the sequence) is within one nucleotide of its comparatively predicted position (X, Y ± 1 or X ± 1,Y).
  9. 8 Accuracy values in bold under the (C) columns for 16S and 23S rRNA represent average prediction accuracies in the current study for just the subset of sequences considered in the previous Gutell Lab studies.[29, 30]. The following sequences were considered in previous Gutell Lab studies, but excluded from the current study, Olisthodiscus luteus (16S rRNA, Chloroplast) and Sulfolobus solfataricus (23S rRNA, Archaea).
  10. 9 When the efn2 re-evaluation and re-ordering step was omitted from our study, the average prediction accuracy was 40 ± 13 for 16S rRNA, 40 ± 13 for 23S rRNA, 69 ± 24 for 5S rRNA, and 66 ± 24 for tRNA. For complete details, see our website[36].
  11. 10 Accuracy scores for the best and worst predicted structures in each group.
  12. 11 Percentage of predicted structures with an accuracy of 20% or less.
  13. 12 Percentage of predicted structures with an accuracy of 60% or higher.
  14. 13 Percentage of predicted structures with an accuracy between 20% and 60%.