Table 2. Sensitivities (sens) and positive predictive values (ppv) for secondary structure models as a function of included information.
RNA | (1) Individual models based on sequence alone | (2) Consensus models including sequence-only alignments | (3) Individual models based on SHAPE reactivities alone | (4) SHAPE-directed alignment consensus models, pairs with >95% pairing probability | (5) Individual models constrained by consensus pairs from SHAPE-directed alignment | (6) Individual models constrained by consensus pairs, structures with incompatible SHAPE data omitted | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
sens | ppv | sens | ppv | sens | ppv | sens | ppv | sens | ppv | sens | ppv | |
E. coli 16S | 60.0 | 52.8 | 66.5 | 68.8 | 90.0 | 84.0 | 76.8 | 91.7 | 93.8 | 89.0 | 96.8 | 89.0 |
C. difficile 16S | 58.7 | 51.8 | 70.9 | 70.1 | 89.7 | 83.8 | 79.6 | 90.4 | 92.7 | 86.6 | 95.6 | 86.6 |
H. volcanii 16S | 78.1 | 68.3 | 59.3 | 69.1 | 89.3 | 81.5 | 74.9 | 91.2 | 90.2 | 85.3 | 92.2 | 85.3 |
E. coli 23S | 68.7 | 59.8 | 60.0 | 64.1 | 85.6 | 78.3 | 79.0 | 89.1 | 90.7 | 83.1 |
All values are given as percentages. Consensus structures were generated by RNAalifold [21] using SHAPE-directed sequence alignments (column pair 4). These structures were then used to constrain individual structure predictions (column pairs 5 and 6). Models were compared against three control predictions: individual predictions based on sequence alone (column 1), consensus predictions based on sequence-only alignment (column 2), and individual predictions made using only SHAPE reactivities and no sequence alignment information (column 3). Covariation structures are not available for C. difficile or H. volcanii 23S rRNAs.