Fig 1.
Two structural states of the thiamine pyrophosphate (TPP) riboswitch in Bacillus subtilis. [18].
A) Riboswitch when TPP is present. TPP ligand approximate location shown as solid circle. Thi-Box and terminator are formed. B) Riboswitch when TPP is absent. Anti-terminator is formed. The aptamer region is roughly between 1nt-120nt and the expression platform is roughly between 120nt-190nt (A and B). C) Folding energy of the TPP riboswitch. The unbound state is more stable, when the elongation is 170nt; the bound state, when the elongation increases to 190nt.
Fig 2.
Prediction of the alternative structure of tenA TPP riboswitch.
Sequence length in the Barsacchi Dataset was 190nt. A) The actual alternative structure S2 (here, the unbound state), with the approximate locations of the aptamer substructure and the anti-terminator stem as indicated. B) The CP-predicted alternative structure . The stem seed L* = {28…34·178…184} is shown in red. C) The SC-predicted alternative structure
(See Materials and methods for details).
Fig 3.
The open blue square represents the computed MFE structure ; the open red circle, the actual alternative structure S2; the solid red diamond, the CP-predicted alternative structure
; and the open red square, the SC-predicted alternative structure
. Blue and red dots represent samples of the MFE-containing and alternative clusters, respectively (see the Materials and methods section for details on our SC implementation). A) A standard depiction of an energy landscape, where the MFE structure
provides the reference structure at distance 0. B) The Alternative-Structure-Referenced (ASR) energy landscape, where the actual alternative structure S2 provides the reference structure at distance 0. Some important base-pair Hamming distances can be read directly from Fig 2A and 2B:
(from both 2A and 2B);
and
(from 2A).
and
(from 2B).
Table 1.
Alternative-structure prediction performance in the Barsacchi Structural dataset.
Fig 4.
ROC. curves for SC- and CP-based riboswitch classifiers.
The blue and red lines correspond to SC- and CP-based classifiers. The legend gives the ROC value, the area under the curves for the two classification methods. Calculations were done at 37°C for all three classifiers.
Fig 5.
Alternative structure prediction in the purine riboswitch dataset.
We examined the predicted structures ,
, and
for Stem P1, a substructure of the bound aptamer. A) The number of predicted structures containing Stem P1. The left bar corresponds to SC (which predicts
and
); the right bar, to CP (which predicts
and
). B) The location of the stem seed L* for the 24 sequences where CP predicted the bound aptamer in
but not
. C) The schematic structure of a typical ligand-bound purine riboswitch.
Fig 6.
Alternative-Structure-Referenced (ASR) energy landscape of xpt Guanine riboswitch.
The sequence length was 162nt, directly from the Barsacchi dataset. In each subfigure, the open red circle on the left represents S2. In Fig 5B–5D, the solid diamond represents CP prediction under default parameters; and the empty diamonds, CP prediction under non-default parameters. A) The solid red diamond represents the CP-predicted alternative structure ; and the open red square, the SC-predicted alternative structure
. Blue and red dots represent samples of the MFE-containing and alternative clusters, respectively (see Materials and methods for details on the Sampling-Clustering procedure). B) CP-predicted alternative structures
under different dissimilarity thresholds τ = 0…60. If predictions were identical, they are shown with a label indicating a range for τ. C) CP-predicted alternative structures
under different temperatures T from 37°C to approximate melting temperature 70°C. The free energy of any fixed structure varies with temperature, and the underlying RNA structure remained fixed for temperatures 37≤T≤49, 50≤T≤58, and 61≤T≤67. D) CP-predicted alternative structures
under different seeds and iterations. The initial number in the label of each point gives the iteration i = 2…5. The seed was either a stem seed L*, a base-pair seed (the base pair with highest conditional probability, denoted by “BP”), or all base pairs with conditional probability higher than a threshold (either 0.5 or 0.8). For iteration number i = 2, the CP-predicted
from stem and base-pair seeds were identical: hence, its label “2-(stem,BP)-seed”.