We are very excited about the results which have been posted in Eterna:
and the data can be found in
Google spreadsheets: FMN/MS2 Riboswitch Structure
or the Excel file: FMN/MS2 Riboswitch Structure
I have data for the inverted same state labs as well and it look good. However, I did not get many "Bad" designs submitted due to issues. I do have alot of high scores in the good groups and some bad as well. Sara scored 100's so that is really awesome considering Sara only did mods and predictions of those mods on untested designs. I will post more as I figure it out. Most importantly though, none of the bad ones scored above 80 as predicted in SSNG1!!!!
Also, something to point out is that in the predicted good there are only 6 out of 49 designs that scored under 60 in the good group which means that the 30 60 groups have been ignored. If you look at the predicted bad there is a break in the middle around 40. That actually coincides with the range I used in DPAT to get the markers for those. I use 20-40 and 50-70 as the score range when choosing 30 60 group markers and the groups fit that very well except for that one little bugger just under 50.
Optimal Small Loop Binding Site Stacks
2ndState-17:35,18:34,19:33,20:32,21:31,22:30,23:29,24:28 MinProb 0.80 MaxProb 0.99
1stState-17:35,18:34,19:33,20:32,21:31,22:30,23:29,24:28 MinProb 0.70 MaxProb 0.93
Optimal Large Loop Binding Site Stacks
2ndState-7:46,8:45,9:44,10:43,9:42 MinProb 0.6 MaxProb 0.92
And here is plot.
And here is the data for Same State NG 2 which also only had good predictions submitted. It also has a very wide range and contains very high scores with 100's.
The predictions were done using partition function ranges, markers and stacks list from DPAT.
I got a couple files mixed up and I am unsure which one is the source file now for the stack pairing probabilities so I don't have the list :(. They are off by a couple stacks though. I ran 2 different experiments and used the same source file for each so there are 2 with similar names. In future predictions I have used a single source folder for Sara. I am also going to code up a report file that gives what settings were used so this is avoided in the future. Sara also used old code for this one in that she only looked for at least on leg in the list of stacks to be present.
I think that based on this data for predictions based on optimal binding site lengths shows promise in that Sara picked out scores in the 100's. The old code/methodology however hampered the investigation I think. I realized this early on and made predictions for what designs would score good and bad in the forums (with controls) using new updated code that could search for an entire lists contents to be present or not present. The only drawback is that the sample size is small so not sure how much good the data will be.
Salish’s end bit discovery put in perspective
I was wondering how the PWKR designs were behaving this round, due to the mystery they caused me in the last round of Riboswitches on a chip. These designs were having extreme high fold change and seemingly a good deal of it was being triggered by just shifting the bases in their colorful tails.
So I went looking in the new Same State NG 2 lab results - really all I had to do to find them was sort by highest fold change. :) Since they were exclusively among the top. Just like last time.
Winner by Jennifer Pearl’s SaraBot (congrats!), score 100%, fold change 154.76, tail bases highlighted.
I took a look on them and the pattern I noticed in PWKR's designs last round, popped up again. (Solely shifting around their tail bases) And as last the UU tail base designs were among the best. Also these UU tail designs were having a fairly low fold change error rate, where most of the others have a slightly higher one. (My current personal preference is anything below 1.15, while I will look at anything that has below 1.2, unless there isn't much good data to choose from.)
I noticed something else. The designs that generally get best fold change, are the ones that have tails bases that repel, bases that can not pair with each other.
There is an extra twist to these top scoring designs, both the absolute high scorer in this lab by SaraBot, the designs by Salish and the original by PWKR. The odd tail pattern was combined with an AU base pair for closing base pair. Similar the other end of the stem was closed by an AU at the FMN aptamer end.
Winner by Salish, score 100, fold change 138.81
I went to bed as it was time. But this pattern woke me up tonight. I think I know what is going on.
Robot necks on human designs
Back in the early Eterna days when we were solving single state lab puzzles, we had big trouble getting the neck area sticking together. However while NUPACK was in trouble in other parts of its designs, it was making a bunch of beautiful working necks. Some of us decided to borrow those fine robot necks and stick them unto the player designs that just had broken necks. And it was working - Aldo demonstrated it could be done successfully.
Example of design with robot neck (orange box) by Aldo
Image screen napped from my post on Energy in the neck area
These robot necks had some pretty unusual patterns compared to what we would normally use for stems. They were rather weak, like only a GC pair at either end holding the stem together, with all AU base pairs in between and them being placed in a repetitive manner.
This convinced me that the neck area was different to the normal stem area. (Later we didn’t have far as much trouble making necks - but it is still the stem area where most deviation from even energy distribution is allowed.
Neck area in switches versus static designs
The neck in PWKR, shows the opposite pattern in the neck than that in a static design. Despite the stem itself should be static, it doesn’t hold GC closing base pairs which is otherwise normal solving pattern (for small designs) and for stems meant to be static.
What I think this does is to allow the stem to flex more than if it have had the GC base pairs at end. I think it goes more flexible and as such is allowing for more moving around inside the design. In a static design things shouldn’t move around too much, it is better if it is stiff. However I suspect that too stiff in a switch design could cost some flexibility.
Basically I think that the neck in static designs, may relieve stress in the middle of the neck (sometimes necks in static designs can also be all GC's and in that case they keep the pressure inside the design in check). Where I think the neck in switch designs may relieve design stress at its ends.
Salish's endbit investigation causing cool lab patterns
Salish’s comment on his PWKR winning design on End bit investigation and eternacon presentation sent me digging in the WIKI. Turns on he has been onto this tail bit pattern for a long time. (See page 8)
I also think I know why this tail base pattern seems more beneficial in the Riboswitch on a chip labs, but not in the microRNA’s. The microRNA labs tends to have dangles, and one or more of the tails are actively involved in the switch area. The microRNA designs don’t need to get internal stem released to the same degree as they are moving in themselves, and not totally locked up in a switch bubble, with only the choice of moving internally.
Whereas the main part of the good Riboswitches on a chip designs tends to have a static neck. I think the tail bases is only having an effect if it is actually next to the static stem.
Why can small tail bits have long range influence?
So why may these odd switch neck ends and its tail bit bases have influence on the fold change and how well the design switch?
The PWKR designs are very actively switching designs, only having a stable neck, but having the rest of the design moving. I think this puts a lot of pressure on the neck. Meaning that it would benefit it to be a little flexible in the ends, to allow for more flexibility on the inside.
A while after realizing that the neck area was different in static designs, I decided that I think there is a particular reason why necks are different to normal stem and why what is going on in them have long range implications.
Tailbits and mismatches
Really this tail base pattern have a lot in common with the mismatch pattern that is aiding stems in switching. Just here the stem isn’t to switch, just have flexibility.
Double mismatch example - put around a switching stem
Image screenshot from this forum post dealing with the topic
Switch neck and tail bit perspective
I expect this tail base pattern to mostly benefit very active switching designs, that needs to let a little steam out, to get the rest moving.
If this really so, I will expect JL’s sliding designs to benefit from a similar behavior. I wonder if it will also have a positive effect on my more immobilized partial moving switches with static stems.
I have been saying that the neck is a special area - apparently that also seems to count for the neck area in switches too. :)
I think this is worth looking out for. It may not be usable everywhere and not for all kind of switches. But wherever it will be of use, it will be nice to see.
Salish, cudos for being the first to notice and propagate this end bit pattern and and for keeping at investigating! Plus thx to PWKR for making awesome weird designs. I still want to know if the double FMN aptamer in his designs can really catch two FMN molecules... :)
Background documents on the Neck area
R101 FMN MS2 Riboswitch Structure, with Switch Graphs
To get the most out of my lab slots I pulled out a few of the absolute best designs and tried a battery of different experiments on them - including rotating them onto other labs when possible (Having the aptamer being rotated the same way in relation to the design) and then make the same experiments again. A good deal of these original good designs, I resubmitted to as for control, for comparison with their new siblings.
My raid of targeted destruction of key elements in winning designs generally succeeded. And in the most obvious case where it didn’t, I’m particularly pleased. :)
The case where my deletion raid didn’t fully succeed, was my quest to find out if the static stem was really needed in the switching area, as I have been claiming and been wondering about. (Search for the section No static stem)
I still haven’t watched all designs, but I will put up my main conclusions here but use this document to fill in the details on the individual labs and for updating along the way.
The static stem in the switching area
Here is the showcase for the question about the static stem in the switching area. First the original design, with a static stem targeted for destruction.
Winner from last round - rerun with same score in this round.
#InverseDoubleMagnetSystem - JL SNG1 3.00 18, Same State NG 2, original score 100%, score 100%, fc (fold change) 27.57
The same design with the static stem deleted
#StaticStemDeletion - JL SNG1 3.00 18 - variant 3, score 100%, fc 28.76
And we have a winner! :)
A full deletion of the static stem in the switching area is possible without adverse effect. Also it is possible making winners with fewer bases left in the switch bubble. Which is really neat knowing.
This design even has slightly better fold change than the original design including the static stem. So it is possible making winners without that static stem. :) Not bad, despite I have been postulating that deleting the static stem would make things worse. Really, here deleting it, made things better. :)
However it is the exception. Almost any other design, that I deleted stems in, suffered in their score and fold change. Not all suffered a lot.
Even more encouraging, there were other winning designs in some of the Small Loop experiment labs. Not many, but enough to show that it is possible making working switches, without that static stem in the switching area.
The more static of these designs seems to have their own blueprint - not too unlike the original blueprints I proposed - but just without the static stem.
What I find really interesting, is that in many of the better solves, despite the 1 state can take many different shapes, the 2 state is often really well in agreement on the overall structure.
Here state 2 resembling real much what I have seen also be successful in the full size riboswitch on a chip designs. The static stems are missing, but the overall structure and placement of FMN in relation to MS2 is the same. Similar distance is varying a little between designs, but the overall impression is the same.
Enrabe, score 93%, fc 14.61
Even some of the top scorers made by ViennaUTC that are full or close to full moving switches comply to this blueprint when looking at their structure in the 2 state. Sometimes with a little bending.
Winner, score 95%
Some skewing of the axis between the MS2 and the FMN are allowed though. But mainly in the small and big switch bubble design, the Small loop Experiment and the more full moving designs as that of PWKR.
So good thing, I still think there is something like an overall blueprint for getting an easy road to a working switch. But this blueprint will vary slightly with size of the design. What to choose will depend on what kind of problem one has.
Fold change versus switch bubble size
Drawings of all the majority types among the winners in Same State NG 2. All of these are winners, but each type achieve a very different max fold change. The PWKR and Xeonanis type are the most prevalent.
These major design types among the Same State NG 2 winners follow same basic overall blueprint for these particular elements - the aptamer will be generally lining up on the same axis as the MS2. With some skewing in the PWKR case.
The blueprint of the 2 state is basically the same, if everything but the MS2, FMN and neck is ignored. What will differ a lot is the 1 state.
So what this image and lab results show is that the more degree of freedom is allowed within the switch, the higher the fold change. All of these designs are winners. But it seems that the size of the switching area will partly determine how high the fold change can go. So now I wonder what this will mean for practical implications.
Function of the static stem
Do I still think the static stem in the switching area has a function? Yes, it really do seems to have a function in most of those designs that are already doing well with it - meaning that it mostly can’t be deleted without the design suffering a loss in score and fold change. I only found 1 exception so far and a few designs that wasn’t hurt too bad.
However it really is possible making a smaller version of a switch work, without a static stem in the switching area. Which is really neat knowing and what I was wondering about, despite being a huge fan of having a static stem in the switching area. However I think this size design will not be able to hit as high fold change - due to the reduced size of the switch bubble.
Similar it is possible making far better fold changes by just having the neck static and the rest of the switch moving. (PWKR and jandersonlee’s sliding switches). This kind of switches take more work hitting on as there are more bases up for mutation. However they still carry overall features of the same blueprint. More on this.
So what switch type to pick, will depend on what we need. Do we need extreme good fold changes, then PWKR type and JL’s sliding designs with a big switch bubble area, are the ones to go for. Just like the natural occurring FMN switch is far bigger and has a much more tight bind than the artificial one we are playing with.
Can we live with lower fold changes, then the reduced design without bases for a static stem in the switching area is likely the fastest way hitting on a working switch. These may have more limited range to hit on winners, however it should be fast getting them, as there isn’t much bases to mutate in them. (Micro switch bubble) While some of them have really sweet fold changes, I’m think they can’t go as high as the PWKR designs. They simply don’t have the space to achieve such a fit.
Or the middle size switch bubble with a static stem - that is fairly easy to hit on and also allows for a good fold change. So I basically think what to choose will depend on practical need.
This leaves back that there are several FMN/MS2 blueprints. There simply are different blueprints depending on what size switch one aims for.
Which way do the aptamer turn in relation to the design? That is the question. I have grouped the labs in order of them having the same orientation of the FMN in relation to the switching area. This brings me to these two categories:
Inverted Same State NG 1, Same State NG2 and Inverted Same State NG 3. (FMN twin G’s furthest away from switching area)
Same State NG 1, Inverted Same State NG 2 and Same State NG 3. (FMN twin G’s closest to switching area)
I have taken the two labs in each category that scored best. I have picked Same State NG 2 and Same State NG 1 for thorough analysis. The other labs generally follows the same trends as their sibling labs. Just the scores are worse. Just as in last round Same State NG 2 did absolutely best.
Hashtag experiments - How did it go?
GU Reduction to GC
Hashtag Experiment name: #GuReductionToGC
Removing a GU base pair from the switching area, by mutating the U in the GU to a C. Intention is to showing that the GU has a function. Expectation - lowering of score.
GU reduction to GC result
As expected the GU’s in the switching area in good designs have a function. When designs are reduction from GU to GC it generally always kills fold change and score too.
Gu reduction to AU
Hashtag Experiment name: #GuReductionToAU
Removing a GU base pair from the switching area, by mutating the G in the GU to an A. Intention is to show that GU has a function. Expectation - lowering of score. (I was targeting Aptamer gate GU’s).
GU reduction to AU result
As expected the GU’s in the switching area has a function. When designs are reduction from GU to AU it often kills fold change and score too.
Especially the Same State NG 2 lab designs, took big hurt. What is really worth noting is that the score in the GU reduced design is often somewhere around 60. Baseline sub score and folding sub score isn’t hurt. But generally the switching sub score is almost totally killed.
The Same State NG 1 and NG 3 designs aren’t hurt nearly as bad by a GU deletion in the switching area. It seems as if the orientation of the FMN matters for how much a GU deletion hurts. In one case score was the same as the original and in a few cases score was not reduced much.
I mentioned beforehand that I suspected that designs with complementary solving style, could tolerate a GU deletion better. The designs that has FMN orientated such that the FMN has its twin G’s furthest away from the switching area, tends to benefit from magnet system style and be more hurt by GU reduction.
Total sum up on change in GU reduction
This was a demonstration that even a single base change can totally kill a design, if an important base is picked.
Not all design types suffered equally from a GU deletion. Designs with FMN’s with twin G’s being close to the design, still mostly did lose score but tended to suffer less.
The GU base pair in the switching area, seems to be filling a function that in particular GC but also often AU can’t. Which is something that I have been after for ages. :)
Static aptamer end deletion
Hashtag Experiment Name: #StaticAptamerEndDeletion
To partially or totally delete the static end of the aptamer. Partial deletion will hurt less than total deletion.
Static Aptamer End deletion - Results
Also as expected, partial deletion hit less hard than total deletion. Partial deletion is not always total detrimental.
The total deletion of static aptamer ends always fare way worse than partial deletion. All designs with deletion of the static end of the aptamer, suffers compared with the originals. Partial deletions also mostly lowered scores - although in a few cases the design was still a winner - due to it being able to accept a somewhat wiggly stem next to the aptamer - provided that the far end was actually static. Omei has earlier noted that there seemed to be a benefit from adding a 1-1 loop in the neck at a fixed distance to the FMN aptamer. See the section: Internal loops and switching. Plus both ends of the aptamer could be moving despite the design was not a full moving switch
I have already been going on about the strange PWKR design with the extreme high fold changes. But I'm far from done.
I was wondering about what was actually considered normal fold change ratios for switches. I found this in a paper:
“The architecture of simple riboswitches composed of a single aptamer and a single expression platform has very limited functional capabilities. Most obvious is the fact that the riboswitch will respond only to its target ligand, or perhaps also to a close chemical analog that may also be present in a cell. Furthermore, the dose-response curve for a simple riboswitch can do no better than conform to the functional optimum for a one-to-one interaction between a receptor and its ligand (Fig. 5). Specifically, a simple riboswitch that functions to perfection will require an 81-fold change in ligand concentration to progress from 10% to 90% gene modulation (Fig. 5B). If some of the riboswitch RNAs being made fail to function properly, because of folding problems for example, then the dynamic range for gene control will be reduced and the dynamic range for ligand sensing may be expanded.”
Not sure how this kind of fold change relates to our kind of fold change.
Ha, I just found something really interesting that may put the PWKR design in right perspective. :)
“The riboswitch has two domains that bind to glycine, while all other riboswitches only have one binding domain, Breaker said. "Binding of one ligand at one binding site improves the binding affinity of the second ligand to the second binding site," he told The Scientist. "This allows the riboswitch to sense much smaller changes in ligand concentration." Most riboswitches will significantly change the expression of the genes over about a 100-fold change in concentration of ligand, but this one requires only about a tenfold change in concentration of the ligand, making it very sensitive, Breaker said.”
Now the glycine tandem aptamer riboswitch has quite another structural outline. See image in this post.
But perhaps the PWKR designs are really catching two FMN molecules at the time. As it really do seem to have a double aptamer domain.
I think a way to tell indirectly, is if the unlocked FMN aptamer is untouched (not mutated) in the designs with the absolutely highest fold changes.
Hmm, now I wonder why some of true FMN sequence got mutated too. Mubot, I really like you... :)
Some of the designs actually were working fine, even with the original aptamer broken. :) Likely showing that the fake extra aptamer can do a rescue mission.
Mutated aptamer - that C should have been a G.
This one got a 100% score. Now I wonder what's it fold change is. Only a 37.12 in fold change, which is unusually low for a PWKR design. They go to the high end 155 in fold change in this round.
That's interesting as this may suggest that the design is actually using both its FMN aptamers. And that one being broken is affecting the fold change.
I wonder if this pattern is general for the PWKR designs that has the true aptamer mutated?
I sorted after fold change in the spreadsheet and the sibling designs with highest fold change, have both the original FMN and the extra FMN intact and unmutated, whereas this is not as much the case for the designs that score lower and have lower fold change. They regularly have one of the FMN's mutated.
Perhaps it wasn’t just only a good laugh, when my niece decided to improve on my RNA drawings by adding in extra aptamers and stems between them in line after the first one. She asked me afterwards if I didn’t think they had become much better now. :)
I decided to continue the fun. If it can work in one type of designs, as it seemingly does, I see no reason why it can’t work in another.
I drew an Exclusion design with an extra aptamer too. I’m aware that the balance between FMN and MS2 would shift in case such a design were made so the structure would likely have to change a bit.
That extra FMN should even give the 2 state some help shutting that MS2 off. :)
I also did a Small loop experiment drawing - including a double aptamer.
And so on. :)
I only have one last naughty question. Can PWKR’s design be made to work with a triple FMN aptamer also? ;)
I am have more data to post but wanted to get this out there as quick since I have alot to do today. I am going to post data for EXNG3 next and then look at data for the static small loops to see if the data agree's with Eli's thoughts on the static small loop which is similar to what DPAT has been seeing.
Here is data for EXNG3 with good and bad large loop binding site predictions. Since the bad prediction actually happened to be the control group (3 base pair long) I was using for everything I ended up using the predictions for static long loop without probabilities as the control. This basically includes all variations of the large loop submitted. It was not the best but it did not think of a control to use back then and just figured I could use it as a control now but not sure if it is a valid control since there are so many different things going on in that one it doesn't really say anything other than there are only 2 groups a good and a bad which I predicted. There were a couple OK designs in the predicted bad group but none that go higher than the high 70's and it seams to focus on the 60 group. The good group has a few where the predicted good's go up to 80 but with the distribution between 50 and 80. The bad predicted group score significantly lower than the good predicted group with its distribution between 10 and 40 mostly. The highest good is not the highest design in the round but it is high.
I am starting to believe that there are a series of ideal binding site lengths and not just a single one which is in agreement with Eli's observations about the small loop binding site. He observed that the longer static loop is better but that the occasional non-static loop could still work thus there is not just a single ideal structure. I have not looked at my predictions for dynamic small loop vrs long static small loop but I will post that data next.
Here is the plot for EXNG3
In this round we have tried out both orientation of the FMN on each lab.
I generally prefer the FMN with the rotation as the left one in the image below. (Same orientation as turn up in Same State NG 2 and Exclusion NG 2.)
In earlier rounds I saw us have trouble with FMN as orientated in Exclusion NG 1 and Exclusion NG 3. I predicted (see the section: Test of the blueprint) that the new labs that we got with the FMN orientated as what I call best, would likely do better with this orientation.
I was right for the Exclusion part.
The Inverted Exclusion NG 1 and Inverted Exclusion NG 3 labs really did show good promise compared to the original labs (Original ones Exclusion NG 1 and Exclusion NG 3). That is even in the first run of them and against labs that we have been struggling with for many rounds. Showing that in some cases that for certain rotation of the main design, a specific rotation of the aptamer is more beneficial.
Which labs got the FMN I thought less good? That would be Inverted Exclusion NG 2 and Inverted Same State NG 2. Did they do worse?
For Inverted Exclusion NG 2, highest score was 86%. Only a few designs scored in the 80’ies. So it do look as if it was harder. We had a row of winners in the Exclusion NG 2 lab, and ever since Perushevs design we have had no trouble getting winners in this lab.
However it was possible to do good in the Inverted Same State NG 2 which got a winner and some designs in the 90’ties. As good solutions from Same State NG 1 and Same State NG 3 with the “worse” FMN’s, could be rotated onto it.
As I have pointed out earlier, generally the Same state labs are less hurt by FMN orientation than the Exclusion labs. When the design is rotated to the less optimal sides - middle is best - they actually in some case they seems to benefit from it.
The Same State NG 1 and Same State NG 3 labs did well thanks to a a special trick, called Hidden hairpin in a loop. And as such it was more than possible to revert the potential damage done by FMN orientation.
What about the Same State labs that got the better FMN orientation that they didn’t have before? That would be the Inverted Same State NG 1 and Inverted Same State NG 3. Those labs didn’t exactly do better, despite I rotated good scoring NG 2 solves onto them.
Particularly in the Inverted Same State NG 1, the rotated designs didn’t do well. Some of the rotated Xeonanis designs did well in the Inverted Same State NG 3.
Score 92%, by Malcolm, rotated Xeonanis type, Inverted Same State NG 3
So while it is possible to rotate both the FMN and some of the Same State NG 2 designs, the middle Same State NG 2 lab likely has the most beneficial rotation achieved yet. For both for best FMN orientation and for best design rotation.
FMN orientation dependent patterns
What becomes even clearer with the new data is that the FMN orientation itself is causing particular solving styles in both Exclusion but especially the Same State labs.
Some patterns are FMN orientation dependent. Meaning they will only turn up in a lab with a FMN of a specific orientation.
Best FMN - with twin G’s furthest from switching area
In Exclusion labs, this calls for pyrimidine MS2 turnoff - targeting of the twin G’s
In same state labs, this encourages magnet stem segments in the aptamer gate.
Worst FMN - with twin G’s closest to switching area
In Same State labs this encourages complementary style
In Same State labs, this orientation encourages Hidden hairpin in a loop patterns.
In Exclusion labs tends to provoke open ended designs, if MS2 is close to RNA sequence ends.
- In both Exclusion labs and but especially the Same State labs, it tends to cause slightly longer aptamer gates.
Conclusion on FMN orientation
Generally it seems most beneficial to have FMN with an orientation that keeps its twin G’s furthest away from the switching area. Unless you are an open ended design.
However if you are a same state lab design, with your MS2 rotated so it is close to either sequence end, you can pull the Hidden hairpin in a loop trick that will make the otherwise harder FMN orientation beneficial.
Hashtag experiment name: #BreakingHiddenHairpinInALoop
In the Same State NG 1 and 3 designs a special pattern were present in the winners. I called it Hidden Hairpin in a loop.
The hidden hairpin in a loop is a FMN orientated pattern. Same state labs that has this FMN orientation and belongs to this group:
Same State NG 1, Inverted Same State NG 2 and Same State NG 3. (FMN twin G’s closest to switching area)
The pattern it is a tetra loop closed with a GC pair - that is turning in a specific orientation - but instead of being placed in a stem, it is hanging in a multiloop area - can be internal loop area also - in one state before forming a stem in the other state.
This pattern, I see as fundamental to the switching in designs of this type. It's a driving force for making the switch happen. It turns up in combination with the complementary style of solving. Really both these patterns are provoked by the FMN aptamer being reversed in orientation towards the switch design, compared to the Same Same State lab and its rotated siblings, Inversed Same State NG 1 and Inversed Same State NG 3.
Both patterns shown in combination
Breaking Hidden hairpin in a loop Results
All the designs where I either removed one, two of these hidden closing hairpin loop bases or reverted them, suffered a rather big score loss. It’s a kill switch on the fold change.
All these labs, have either winners or topscorers carrying this pattern. There were a few topscorers in the new lab we only tried out this round, Inverted Same State NG 2, although it were possible solving in alternative manner also.
Patterns that works for this type lab
There is an orientation to the the base pair making up the hidden hairpin in a loop. When reversed, the lab designs loose score.
NB, for the Inverted Same State NG 2 lab there is only wittle winning data. But I expect the pattern to hold true also there, if further tested.
Breaking Hidden hairpin in a loop background
Same State NG 1
Score 100%, mod by SaraBot, design with hidden hairpin in a loop highlighted
Only the bottom one is missing one of the bases in the hairpin and it is the lowest scoring of these designs.
Same State NG 3
Score 95%, jandersonlee mod
Hidden hairpin loop pattern highlighted in high scoring designs
Inverted Same State NG 2
Here we need to be looking for two kind of hidden hairpin in a loop pattern as both the Same State NG 1 and Same State NG 3 is possible.
Score 92%, I had rotated one of Mats Same State NG 3 winning mods onto the new lab.
SSNG 3 type
SSNG 1 type
So as can be seen these pattern are there among the high scorers, but they don’t stand strong out yet. However I think they will in case of more runs.
Why rotate labs and was the result?
What the rotated designs have shown us is that winning designs can be rotated on to a differently orientated lab (with same FMN orientation) and will often still score fairly well. In most cases the newly rotated design don’t score better than their original.
But the sweet promise of rotation is that by making rotatable designs, one may achieve an even higher fold change for one of the rotation.
Switch size, number of static stems and rotability
The number of static stems affect rotability. The fewer static stems, the less rotability.
Small switches will be likely to carry a similar blueprint too - just without the static stem. The lack of a static stem in the switching area cost them their rotatability.
Small Loop Blueprint
Similar open ended switches with just one static stem in the switching area - as tend to turn up for the Exclusion NG 1 and Exclusion NG 3 - that have the twin G’s in the FMN facing towards the switching area, are non superimposable and thus not rotatable either.
Middle size switches with static stem/s in the switching area - rotatable - on 2 or 3 axis depending on number of static stems. Will also tend towards carrying a particular blueprint
Big switches - like PWKR’s and jandersonlee’s gliding switch are well possible. They typically only have one static stem - the neck. Their blueprints may be more variable and harder to predict. Due to lack of static stem in the switching area, they are not rotatable.
Perspective on rotation
The promise of rotatability is the chance to hit on an even better fold change and score with one of the rotations. For now it seems that we already hit on the optimal orientations in the original labs we had for the two first Riboswitch on a chip labs. While it was also well possible making winners for a good number of the new labs.
However as I also demonstrated in one case, rotating a design and then mutating it slightly to fit its new rotation better, can improve score and fit the design better to its task.
Which is why I still believe that new and rotated Inverted Same State NG 2 lab may achieve an even higher score than the Same State NG 3 lab, that for now seems to be the highest scoring of all the Same State NG labs with the FMN rotated to have its twin G’s closest to the switching area.
So I still keep my mind open to the Inverted Same State NG 2, for making an even better results. It was only first run of this labs, and while we got a winner, I say we haven’t seen its full potential yet.
This lab has inbuilt an extra bonus. Just as the Same State NG 2 lab - carries the option of allowing for both Exclusion 2 and Exclusion 3 solving patterns, similar the Inverted Same State NG 2 lab allows for both Same State NG 1 and Same State NG 3 solves. Plus the Same State NG 2 seems to benefit from letting steam out of the neck - being the static end of the FMN and being attached to the FMN directly. Something that the Same State NG 3 doesn’t.
As I mentioned earlier, the L7AE hairpin can open up at both ends. Where MS2 only opens at one end. This means that the L7AE hairpin promises double as many open doors for rotation as the MS2.
I have illustrated this in a Xeonanis style design that has two static stems, beside the neck and as such has 3 axis for rotation as the number of static stems determines the amount of axis for rotation. That is unless the switch element hairpin, like LAE7, can open up itself. MS2 can’t and as such only leaves the xeonanis puzzle with 3 axis for rotation.
Xeonanis type design with 3 axis of rotation, getting a 4th with L7AE instead of MS2
As mentioned by Eli above, refer to slie 8 in this file for details.
One hypothesis is that these endbits do not have considerable influence, since they barely affect the molecule and are not actively involved in switching or binding. A few exceptions exists when, out of a set of 16 possibilities a few (typically 1-4 of a subset with one particular binding partner at either end) actively bind into the middle of the molecule, and completely change the structre. Previous best scorers in ESc can very well end up with 0 ESc as a result. These shall be excluded here. One example is shown on slide 12 in the file above.
But now to R101:
On the example of ExNG1, the cluster density is not dependent on the SSc scoring. This is excellent news, especially since clusters are all on the order of >200, so the data are relevant for our stats analysis, and we can derive some meaningful conclusions from them.
Then, we look at the effect on the endbits onto the scoring. The original design is the third best scorer at with a CC endbit at Pos (1:84). I developed two designs that scored even higher, namely UC (7 % higher score), and AC (20 % higher score).
Interestingly, the dependency of the entire molecule, all other things being equal, on this tiny last base pair is amazingly clear from the data.
Let's take a look at what the data say:
Here the dependence of the SSc on the first and last base, at Pos (1:84). Essentially, score range from "outstanding scorer, exceeding the originally set maximum score of 30" to "barely even switches well at all", all as a result of the changes in these endbits.
Now, if we further analyze these endbits and simply categorize them as "binding" as defined by their joint affinity for one another and "unbinding" as in typically do not bind (caveat: In the past, we ran some labs, where we tested the hypothesis that sometimes UC could be potential binding partners - this will be treated as non-binding here, though), we derive at this astonishingly clearly distinguished separation for the Ex NG1:
Preliminary conclusion: Independent of the actual binding energy of the Pos (1:84) bases, the ExNG1 molecule significantly benefits from having non-binding endbits, es expressed in high switching scores.
Hypothesis: Influence of endbits is not as important as in ExNG1
We still have a significant step-change in SSNG1 to the maximum fold change.
However, the effect of non-binding base pairs seems to be specific to ExNG1.
Great work with the graphs.
You might want to reconsider Exclusion NG 2 as benefiting from the end bits. We haven’t had too many winners in this lab earlier and I see the main part of the winners carrying one kind of tail bit.
I have counted in the GU base pair. It is a non Watson Crick pair, generally weaker compared to both GC and AU. Its called a wobble base pair. Tending to have a wider distance between its bases.
I did a highlight of the designs where I would count your end bits helpful in.
Of Watson Crick end bits, there are only one GC tail bit among these high scorers. AU is the most prevalent. AU is weaker than GC and normally most necks in static stems would prefer having a GC closing, unless pressured to accept something else. It is possible if just one compensate and strengthens other bits. But GC’s are preferred.
You have only two cases of GC pairs at top of your design (1-84) where there is an identical design with different end bits.
They score fine, but in both cases there are higher scoring designs with non binding pairs. Though not by much in the second case.
Score 93%, fold change 18.32, fold change error 1.11
Score 98%, fold change 24.93, fold change error 1.10ttp://www.eternagame.org/game/browse/6369184/?filter1_arg2=6415481&filter1=Id&filter1_arg1=6415481
Score 98% Fold change 25.43, fold change error 1.18
Score 99%, Fold change: 25.43, fold change error 1.14
Interestingly enough all energy models show this GU end bits as split. Usually GU’s are shown as pair up in the energy model.
FMN - forward & reverse - orientation matters
I have been drawing up trends for Same State and Exclusion lab to illustrate that the orientation of the FMN matters quite a deal. It matters a great deal for several reasons.
Most of this I have already tried explain with words earlier, but I find drawings sum things up better. Still a lot of those word thingies along. Sorry! :)
Riboswitch on a chip labs versus NG labs
Our first many riboswitch on a chip labs had the FMN turn in the same direction when it came to sequence order 5’ to 3’. That mean that no matter which end you start from in the RNA sequence, the FMN aptamer would always be orientated the same way. In this case with the FMN twin G’s closest to either of the design ends.
But that also meant that the FMN was differently orientated towards the switching area itself between these labs.
In the latest round with the NG labs we tried all the opposite orientations of FMN. I expected the Exclusion labs Inverted Exclusion NG 1 and 3 to gain better scores, as I suspected that designs would benefit from having the FMN turn the same way in relation to the switching area, as the already successful Exclusion NG 2 lab. And indeed they did get better. I could regularly rotate a design with a FMN facing the switching area in the same way, onto the lab where this option was open and have it work relatively fine. But not in all cases.
The Inverted Same State NG 1 and 3 labs however didn’t get better scores as I had hoped. There was more than just FMN rotation towards the switching area in play.
What aspect of FMN orientation matters?
The orientation of the FMN in relation to the sequence (5’ to 3’)
The orientation of the FMN in relation to the switching area
The order of MS2 in relation to FMN
The orientation of the FMN even effected Switch and Exclusion labs differently. Depending on which way the FMN turns and in relation to what it sparks a different set of patterns.
Lab drawing overview
Drawing showing switches ordered after orientation of the FMN in relation to the sequence (5’ to 3’ direction). (Inverted labs versus non inverted labs)
The packmans indicates the more delicious switches. :)
Drawing showing switches ordered after orientation of the FMN in relation to the switching area
Drawing showing Same State designs ordered after orientation of the FMN in relation to the sequence (5’ to 3’ direction). (Inverted labs versus non inverted labs)
Orientation of the FMN in relation to the switching area
In the same state labs, the FMN twin G’s are closest towards the switching area, tends to spark:
Hidden hairpin in a loop pattern
Example labs: Same State NG 1, Same State NG3 and Inverted Same State NG 2
Tending toward fuller moving designs
Example labs: Exclusion NG 1, Exclusion and perhaps Inverted Exclusion NG 3
The order of MS2 in relation to FMN
MS2 before FMN sequence tends to spark:
Open ended designs in exclusion labs
Tending toward full moving designs in
Long aptamer gate
Example labs: Exclusion NG 1, Exclusion NG 3, Inverted Exclusion NG 1 and Inverted Exclusion NG 3.
The orientation of FMN in relation to 5’ to 3’ direction
Long aptamer gates seems to be caused by either
MS2 at end of the RNA design ie not between aptamer sequences
FMN twin G’s closest to the MS2
Example labs: Same State NG 3, Same State NG 1, Exclusion NG 1, Exclusion NG 3, Inverted Same State NG 2 and Inverted Same State NG 1
Sum up of the generally most optimal orientations of FMN
Having MS2 in between FMN sequences
Having the twin G’s in the FMN furthest away from the MS2
Having the twin G’s in the FMN sequence closest to the 5’ or 3’ end.
The Same State labs are most often able to compensate for badly dealt combos. These are turn on labs which is an advantage of its own.
I use the same cut-off I used for my investigation, at salish99_ExNG2_005 #endbit investigation, ESc 92.55 and analyzed the endbits.
WCP/non-WCP's are just at 35 % at 64 WCP's out of the 183 designs, suggesting a strong influence of non-WCP's on improving the efficiency of the design.
So, looking at these data, we find
1) The first set of 37 designs scored 40 in the SSc.
so, we then look at the fold change dependence on Pos(1:84):
Which leads to some interesting sorting - I rearrange the data to sort by fold change
We cannot display where the 35% of WCP's are hidden in the 183 datum points, thus we zoom into the first 20 and we fidn the following:
or, to express this in terms of WCP/Non-WCP's:
If we zoom out, the effect becomes even more emphasized:
So, there is clearly a positive effect of using nWCP endbits on fold change in the ExNG labs.
So, the data in ISSNG3 suggest:
While the hinges have an influence on overall scoring and while they increas A % by about 5 % absolute, the baseline scoring as well as the swtich subscore may be impacted.
In ISSNG3, the hinges had no severe detrimental effect (0 score or near-zero score), in fact, some of the endbit variations resulted in far lower scores than planting a string of 4 A's into the middle of the design.
So, if we look, for comparison, at the entrie set of designs, sorted by ESc:
we see that the score of the designs with quad-A's don't excel, but also do not fail.
Interestingly, adding two hinges in design 6420106 also did not detrimentally affect the RNA performance, as I initially postulated, but this is also just the scond best performing design. Additionally, we discussed whether having a paired quad-U section would aid the hinge in switching. This is not the case int he SSNG3 set, the addition of UUUU to the RNA did not enhace performance.
Also, varying the hinge in between these sort parameters of positioning did have some impact on the RNA performance, but overall, this effect was small. We will have to look into how the performance was changed when the quad-A hinge was introduced outside this central region that already contained an A triplet - we will go to other sub-labs for that.
Comments, as always, are welcome.
This does not come as a surprise, as the molecule must be working well to start with. It does, however, imply that int he case of ISSNG3, the Quad-A hinge does not play a crucial role in improving the switch subscore.
Lately I have been working at summing up my FMN/MS2 blueprints drawings in one. I had drawn expectations for both our original Riboswitch on a chip labs, the new NG labs with inversions included and for the new Small loop experiment.
I think I both for Exclusion and Same state labs can sum the best options up in one image each.
Now with Salish’s End bit bases added too. :)
This does of cause not mean that Exclusion and Same state labs can't be solved in many other kind of ways. As you have all helped show - and thanks! - they very much can. :)
However I think these starting points are what gives these switches the absolute most favorable orientation of its different elements, from what I have learned till now.
I basically think the most important elements are the neck and the switching elements of cause.
The neck is closing up the switch allowing the switching elements inside a reduced room for finding each other.
Similar the position of the switch element in relation to each other decides if the elements enhance or work against each other.
A static stem in the switching area may play a role too, since it in some cases seems it is not (yet) possible achieving as high fold change without it. At least in the exclusion lab.
Special Same State and Exclusion patterns
In Same state designs, the MS2 seems to almost insist on being geometrically placed in the middle so it is on line with the aptamer. This does not mean that there necessarily are same amount of bases on each side. But as long as there is a not too different amount of bases in the multiloop or ring holding the MS2, things seems good.
In Exclusion designs, the MS2 overall prefers being placed real close to the last part of the FMN sequence. This is a pattern that hit through from the 4'th round of Riboswitch on a chip lab. A MS2 position pattern that Perushev made work in the Exclusion 2 lab that we had a much harder time making winners in compared to the Exclusion 3 lab. This MS2 position since took over everything in relation to the Exclusion NG 2 lab. It is possible making winners with an early positioned MS2 but they are very rare, telling they are much harder achieving.
This MS2 next to the last FMN sequence pattern even hit strongly through in the far smaller Small loop experiment designs.
Put red arrows on to highlight the MS2 position in the Small Loop of the Exclusion kind.
Image taken from my Small loop blueprints post
Early blueprint drawings taken from the original Riboswitch on a chip labs
Recent blueprints for this NG round
I then looked through my data from the last 2 rounds and the best small loop formations are the long small loop. I am starting to think that a completly static small loop may not be "best" for the EXNG1 labs but it definitly needs to be longish. You can see that there is a 93 score for stack 2ndState-18:36,19:35,20:34,21:33,22:32,23:31 but that 93 is not in the 1st state which points to a non-static small loop.