Commit 7ad05b4e authored by Vladimir Reinharz's avatar Vladimir Reinharz
Browse files

FigureS9+ results v3.0

parent 09b4f07a
......@@ -488,8 +488,8 @@ Thus, for each \shape profile distance measure, each \shape distance threshold $
\section{Results}
\label{sec:results}
We evaluated \soft on a comprehensive set of values for $\delta$ the \shape profile distance measure and $\gamma$ the proximity threshold. For each $(\delta,\gamma)$ pair, a Receiver operating characteristic (ROC) curve was computed and the Area Under the Curve (AUC) for each PDB model was averaged for the 5S rRNA in Fig.~\ref{fig:auc} first row and for c-di-GMP in Fig.~\ref{fig:auc} second row.
It is important to recall that the set of positives and negatives is influenced by the value of $\gamma$, as they are determined by the Algo.~\ref{algo:pos}. Notice also that we only show the pair of parameters $\delta, \gamma$ such that the all structures for each RNA had a set of positives and negatives. As discussed before, many nucleotides are missing in the PDB models.
We evaluated \soft on a comprehensive set of values for $\delta$ the \shape profile distance measure and $\gamma$ the proximity threshold. For each $(\delta,\gamma)$ pair, a Receiver operating characteristic (ROC) curve was computed and the Area Under the Curve (AUC){\color{red} are shown in Fig.~\ref{fig:auc}, averaged for each the 5S rRNA and for c-di-GMP riboswitch who have four models.
It is important to recall that the set of positives and negatives is influenced by the value of $\gamma$, as they are determined by the Algo.~\ref{algo:pos}. Notice also that we only show the pair of parameters $\delta, \gamma$ such that the all structures for each RNA had a set of positives and negatives. As discussed before, many nucleotides are missing in the PDB models. of 5S and c-di-GMP.}
\begin{figure*}[ht!]
\centering
......@@ -507,7 +507,7 @@ It is important to recall that the set of positives and negatives is influenced
\label{fig:auc_remu}
\end{subfigure}
}
\caption{{\bf Overall performances of \soft using experimental and computationally-predicted structural disruption data.} For a set of extreme percentile cutoff of the \shape profile disruption in the first {\color{red} row} (computational \remu disruption in the second {\color{red} row $\delta$ and a minimal distance $\gamma$ from the mutation we show the average AUC.
\caption{{\bf Overall performances of \soft using experimental and computationally-predicted structural disruption data.} For a set of extreme percentile cutoff of the \shape profile disruption in the first {\color{red} row} (computational \remu disruption in the second {\color{red} row) $\delta$ and a minimal distance $\gamma$ from the mutation we show the average AUC.
5S positive set composed of the binding interfaces with other chains present in its four PDB models.
The tRNA positive set is divided between the anticodon positions and the A-$\psi$-C-G motif positions, obtained from the litterature.
The c-di-GMP, cobalamin and adenine riboswitches positive sets are composed of the positions at most 5$\AA$ from their ligands in their PDB structures. Four different models exist for
......@@ -523,7 +523,18 @@ It is important to recall that the set of positives and negatives is influenced
{\color{red}
The results shown in Fig.~\ref{fig:auc} all follow the same trend
The results shown in Fig.~\ref{fig:auc} show that in all cases the maximal AUCs are found wen the \shape disruption percentile cutoff $\delta$ is around $97\%$, with the maximal distance possible distance $\gamma$ from those mutations.
The 5S RNA and tRNA show a much more diffuse pattern. We hypothesize that the weaker signal is due to the complex
and numerous interactions helping to shape and stabilize their 3D structures which was not captured during the MaM
experiment. Thus creating noise as the disruptive effect of each mutations.
The strong signal of the c-di-GMP riboswitch might be a consequence of a small artificial helix that was introduced to stabilize the crystal structure, and conserved in the MaM experiments.
The cobalamin riboswitch is the only that exhibits a strong negative correlation at a slightly smaller disruption cutoff than the optimal. This region correspond to the selection of the positions in the leftmost hairpin as disruptive, as seen in Fig.~\ref{fig:shape}, and might indicate another binding interface not identified by our approach.
The two adenine riboswitch MaM experiment show that although similar results are achieved, the variation between
experiments remains a concern and some the quality of the \shape experiments must be taken into account. The necessity of the correct structure when applying the MaM protocol is necessary as negative results are obtained when using the non-bonded form (shown in Supp. Mat. FigS9).
}
%The ROC curves for the 4 PDB structures are shown on the same graph, for a specific \shape distance threshold $\delta$ and $\gamma$.
......@@ -542,12 +553,12 @@ The results shown in Fig.~\ref{fig:auc} all follow the same trend
We conjecture that the differences in the influence of the $\gamma$ parameter, minimal distance from the mutation, are due to
structural differences. We present in Fig.~\ref{fig:dist} the distribution of path lengths (distance) between every pair of secondary structure elements, weighted by the number of combinations of positions that are not in the intersection of the secondary structure element. We observe that the 5S rRNA has a much more Normal-like distance distribution, {\color{red} while the c-di-GMP (Fig.~\ref{fig:dist} and cobalamin riboswitch are more Poisson-like. In contrast, the three way junction in the tRNA and adenine riboswitch induce bimodal distributions.
structural differences. We present in Fig.~\ref{fig:dist} the distribution of path lengths (distance) between every pair of secondary structure elements, weighted by the number of combinations of positions that are not in the intersection of the secondary structure element. We observe that the 5S rRNA has a much more Normal-like distance distribution, {\color{red} while the c-di-GMP and cobalamin riboswitch are more Poisson-like. In contrast, the tRNA and adenine riboswitch havemuch more bimodal distributions.
Those distributions determine how smoothly the number of positions considered decreases as the parameter $\gamma$, minimal distance from the mutation, is increased.
}
\begin{figure}[ht!]
\begin{figure}[th!]
\centering
\colorbox{red}{
\includegraphics[width=0.47\textwidth]{Figure6.png}
......
......@@ -224,6 +224,16 @@ The results for c-di-GMP having as positive positions the ones interacting in th
\label{fig:cdigmp_hairpinornot}
\end{figure}
\subsection{adenine MaM without ligand}
We show in Fig.~\ref{fig:adenine_4} the results for the adenine riboswitch when the disruption is based on a MaM experiment in absence of the adenine ligand.
\begin{figure}[ht!]
\centering
\includegraphics[width=\textwidth]{FigureS9}
\caption{AUC results for the adenine riboswitch when the disruption is based on a MaM experiment in absence of the ligand}
\label{fig:adenine_4}
\end{figure}
\section{Dataset binding positions}
\begin{table}[ht!]
......@@ -237,10 +247,7 @@ The results for c-di-GMP having as positive positions the ones interacting in th
tRNA & anticodon & RF00005 & & $34-36$\\
& T-$psi$-C-G & & & $54-57$\\
& Prot. DNA & & 1EHZ & 1, 19, $34-36$, $56- 57$, $73-76$\\ \hline
c-di-GMP ribo. & c-di-GMP & RF01051 & 3IWN & $8-10, 28, 38, 82$\\
& & & 3MUT & $18-20, 38, 48, 92$\\
& & & 3MUV & $18-20, 38, 48, 92$\\
& & & 3MXH & $18-20, 38, 48, 92$\\\hline
c-di-GMP ribo. & c-di-GMP & RF01051 & c-di-GMP pocket & $11-13, 40-41, 83$\\\hline
cobalamin ribo. & B1Z & RF00174 & 4GXY & $41-43,64-66,72-78,106,108-109,124,148-150,155-157,159-162$\\\hline
adenine ribo. & adenine & RF00167 & 1Y26 & $21-22,47,50-52,73-75$\\\hline
glycine ribo. & glycine & RF00504 & 3P49 & $35-39, 46, 48-42, 110-114,137, 139-143$\\
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment