User:ElNando888/Blog/FMN Switch Mimic Pilot 1
Following will be all about the lab titled "FMN Switch Mimic Pilot 1"
Everyone is cordially invited to discuss the topic / ask questions on the discussion page.
(work in progress)
So, what's a "mimic" anyway? In our context, we're simply trying to make a sequence act as if it is the original in presence of FMN, and since we want to do so without changing the environment, in other words, without adding FMN in the mix, then we have no other choice but to do that by means of mutations localized in the binding site.
The above definition is fairly abstract, I'll grant you that. Before talking more about mimics, I'd like to show you something about the "normal" case, which is, when we test switches using actual FMN.
The dotplots of a switch
On the left, you have the dotplot for unbound structure, calculated with Vienna RNA and the 1.8.5 parameters, which should make it very very close to (if not exactly like) the dotplot you would see in EteRNA.
On the right, you will find the dotplot of the bound structure. As you can see, the MFE in the bottom-left corners is different, which is what we would expect from a switch. And other notable differences are that some strongly marked stacks in one plot become blurry in the other, and vice-versa. And this sounds also very reasonable.
The very unfortunate fact about the right-hand plot, is that there are currently no publicly available tool to calculate this one. EteRNA cannot currently compute this plot for the bound structure and show it to you. And neither can the official ViennaRNA servers. We are working on finding a way to make this tool available to the community, but it may take some time before we're capable of offering the service... Of course, we'll keep you posted.
Still, there's a good news: we can now define visually what is a mimic: the dotplot of a (good) mimic should simply be as resemblant as possible with the bound structure dotplot.
Mimics & comparisons
One remark before we start comparing plots.
Because the mimic provides a free energy boost by making pairs inside the binding site, it should not be surprising to find dots in the red marked areas. And this should be ignored here.
|Mimic 1: CCUAUC--GAAGG|
This may look like a very good similitude, but it is noticeable that the shadow of the unbound structure is a lot fainter with the mimic. What does that mean? My interpretation is that the mimic produces a boost that is way too large, and it extinguishes the probabilities of the other suboptimal folds. The trouble with that is that this mimic alone may be telling you that the switching with FMN would occur, but the energy gap is actually too large. In technical terms, this mimic is probably prone to generate false positives.
This is a personal assessment though, I'd be glad to discuss it further on the talk page.
|Mimic 2: GUAGUA--GAAAC|
This is a loop matching one of the consensus sequences for a so-called sarcin-ricin loop. I was hoping that the structural similarities (in 3D) between the FMN-bound form and this other structure could be good enough to create enough boosting... It would seem that the potential for misfolds, and specially the interactions with the 3' tail are probably going to make this mimic fail, possibly spectacularly so.
|Mimic 3: UGUAUU--GAAGG|
The MFE looks correct with the mimic (ignoring the interior of the binding site and its mutated bases), but the probabilities look quite odd, not to mention the 3' tail... The least we can say is that the result is uncertain.
|Mimic 4: AGGAUA--UAAUU|
This looks clearly failed. It would seem, the mimic is predicted to actually boost the unbound state, more than the bound state.
|Mimic 5: UGGAUA--GACGG|
Looks ok with the MFE, but seems imperfect when looking at the general picture of the probabilities distribution.
|Mimic 6: CGUAAC--GACGG|
|Mimic 7: AGUAUA--UAACA|
|Mimic 8: CGUUAC--GGAGG|
Some of you may wonder why we will be testing mimics that seem to blatantly fail.
For one, the case presented above is just one single sequence, one single case. Statistically speaking, these mimics may prove better or worse than shown above. This is one of the reasons why we will be running them against a set of 20 sequences.
The other very important consideration is that all of the above is purely in silico. All of it are just predictions from a model which we know is imperfect. For that reason too, it is important to test as may cases as possible, even though they may seem hopeless or pointless. Also, the imperfections of the model could be the cause of relative miscalculations. Let's imagine for a moment that we know that the model overestimates the stability of multiloops. In our current problem, this has the direct consequence of shifting the domain of the free energy differences that may produce a good switch. Or, trusting this model could lead some players to create designs that later appear to fold nearly exclusively in one form or the other (does that sound familiar to you?)
Also, measuring failures may help us identifying those more clearly and early in the future.
Based on 25 puzzle solutions (not lab submissions)
|Score 10mM||Score 200uM|
A screening tool
Another important fact about the mimics: they can only tell us whether a sequence has a good chance to be in the correct thermodynamic range for a possible switching. It is important to realize that this study will only be the first step towards designing successful switches.
But if we can create a list of demonstrably good switch candidates, while using the Cloud Lab naturally, then we should also enjoy much better success rates when the Das Lab allows us to run these candidates with the real FMN.