Results for Eterna R98: The Real Logic Challenge using NuPACK
I found out that my first post was so seriously flawed that I decided to rewrite it. I got confused by both oligos being called oligo 1 in two states, when they are really different oligos. My bad. :)
Quick overview of the incoming results
I find that there are several structural types that goes again through the different labs.
I have drawn some structural overviews and added below each summary. They are not to be taken fully literal. Loops may be present and stems may be bending, without it fully showing. I try hit on a very rough overview.
There are a good deal of general things, even across labs sometimes. Like there is often one or two 1-1 loop between the microRNA binding with the RNA sequence. The microRNAs still very much like to bind up almost in full except for some end stretches sometimes.
Two purine boosts for switching hairpin stem loops seems to be common. Or other boost combinations that make the end loops go weak. End loops in switching stems, can be made weak by either being larger, by boosts, or by being a tri loop.
Also I got confirmation that small switching stems between MS2 and microRNA landing spots do well with tri loops. I have been suspecting as they should be easier to break compared to tetra loops. Similar bigger end loops are a help for stems in the switching area, that needs to switch.
There are a set of shapes that turn up, that aren’t too different. I find 3 main types among the high scores, two of them strongly related.
The oligos seems to prefer landing on either side of the MS2 in the 4’th state where both oligo bind up to the RNA sequence.
MS2 likes a somewhat middle position. Oligo 1 and 2 can go on either side of the MS2 and still work.
The majority of the top scorers have the strong Oligo 2 after the MS2 and the weaker Oligo 1 before MS2.
Stems are in between the switching elements in some states. Sometimes they are static between up to 3 states.
While microRNA dangles works beautiful for JL’s high scorer, that has the microRNAs at each end of the RNA sequence, the microRNAs seems to want to get somewhat close to the MS2, even if this means dangling tails with just A and U bases - not meant for anything.
The area around the MS2 can be either mainly purine, but also mainly pyrimidine. However purine seems to win out on amount.
This lab seems to be the least picky when it comes to alternative ways of solving.
End loops for when either of the microRNA landing spots are packed away tend to be positive in energy and or fairly big.
Middle stems that between MS2 and microRNA landing spots can have a tendency to be either bigger loops or tri loops, both things I think are helping breaking open the stem if it is partaking in the switching.
3 of the top scorers have two pyrimidine bases in common just around the MS2 hairpin.
In this lab there are fewer structural types. I found 1 main type and a minority one. The one with the Oligo 2 before Oligo 1 in state 4 wins out.
MS2 likes an end position. The MS2 seems to handle fine being close to the end of the sequence - after both the micro RNA's landing spots.
The oligo landing spots are close, when both are landing in 4’th state.
I wonder if the small static stem afterwards is helping or if it can be removed. This lab doesn't have the MS2 held from both ends, This is contrary to earlier Riboswitch on a chip labs. But then again it is a turn on lab, not a turnoff lab as the Exclusion 1 and Exclusion 4 labs from the Riboswitch on a chip series, so MS2 needs to be present most of the time.
Only the a few designs of the minority type that score 80 and 85 have the oligos in the opposite order. Oligo 1 before Oligo 2. The minority type has the MS2 between the oligos just like the AND lab seems to prefer.
The strong Oligo 1 prefers to be before the weaker Oligo 2 and both wants to be before the MS2. Except for the minority type which the oligos landing on either side of the MS2, similar to the AND lab, except here they are like next to the MS2 hairpin, which was absolutely not the case for the AND lab.
Here we don’t have many top scorers, so it is harder to add pointers to what direction things will go.
The structure of state 3 differs quite a lot from state 3 in the AND and OR gate labs. It seems to need some extra bends and pairing.
There seems to be a want for having the MS2 in between oligo landing spots, but while later oligo landing spot is very close after the MS2, the first is very far away.
The stronger oligo 2 landing spots comes before the weaker oligo 1 landing spot. This I find interesting as it is opposite pattern against the AND lab.
The oligo 1 turns off MS2 in state 3. Oligo 1 and 2 share the task with turning MS2 off in state 1.
The AND gate lab tends to have the weakest oligo 1 before MS2 and the strongest oligo 2 after MS2.
The OR gate lab tends to have both oligos before the MS2. With the strongest oligo 2 positioned before the MS2.
The XOR gate lab top scorer has the strongest oligo 2 before MS2 and the weakest oligo 1 after MS2.
I think it will end up with these different types of labs will prefer having the oligos in a particular order, in relation to the MS2 and the oligos themselves. Now the microRNA can change and as such the sequence. However I think that there will still end up being a distinct preference of each the labs for how to catch these oligos.
In some labs it is possible reusing a bit of the same structure or static stem between some of the states. Things worth considering when this will be beneficial to play.
I’m wondering about what role microRNA catching dangles will play in labs with more oligo inputs compared to the single input microRNA labs, where dangles seemed to be almost obligatory. I see several more cases of micro dangles in form of magnet segments.
Amount of states with MS2 turned on and off
Things may of cause change a bit for patterns emerging for the logic gates labs mentioned above, when we get winners, but I don't think for much.
Half loose MS2
In hindsight it makes good sense that a half loose MS2 - that is not held from both sides - is working quite well in the OR lab. In this lab the MS2 needs to be turned on in 3 out of 4 states and as such is in far less need for a turnoff sequence/s.
A MS2 can always gets turned off, even if loose, if there is just a long enough sequence match for it. What usually would be harder in a single string RNA design - is turning the MS2 off - unless the MS2 turned of by a MS2 similar/identical sequence and is competing against two turned on MS2’s. (Background post on Cooperativity lab) However the two oligo inputs much shifts balance in a direction where a long MS2 turnoff is much more favorable. I shall explain this in next coming post below.
I wasn’t not surprised by the AND lab liking having the MS2 between the oligos. Basically what was exploited in the AND lab is that when the two microRNA landing spots are on each side of the MS2, there are always some complementarity in the oligo landing spots that can be used or partially added to, for making them turn off MS2. It's basically an extended version of nearest neighbor strand pairing, with part of the MS2 sequence swapping partner, between pairing with itself and its next door neighbor strand/s.
Similar for the OR lab, the strongest oligo 2 seems to prefer being furthest away from the MS2 when the MS2 needs to turn on, as the MS2 should only be turned off in 1 out of 4 states.
The strongest oligo 2 complement would be most likely also be strong as the oligo 2 as our microRNA results earlier showed that microNRA and RNA design mostly wants a full and strong pairing up. Thus the oligo 2 complement has good potential matching base stretches to the MS2, which is also strong. So having distance between MS2 and the complement of the strong complement can help for not having unwanted action. Avoiding offering nearest neighbour strand opportunities.
So the principle for the OR lab with 3 out of 4 states having MS2 turned off, seems to be, in the state with MS2 turned on, just leave a static stem or a non complementary sequence at the side of the MS2 that is not to be involved in anything, as both the oligos will likely go to the other side of the MS2. And the MS2 will do what it likes most, in the one state where it needs to be on. Pair with itself and its protein molecule.
The XOR lab is kind of a middle case between the OR and AND lab. Half of the states with MS2 turned off and half with MS2 turned on. It shows tendencies for needing the oligos on either side of the MS2 as it would still need to have a MS2 turnoff for the sequence not active when the oligo 1 binds on either side of the MS2. But it shows opposite oligo positioning pattern to the majority patterns for oligo order in both AND gates and OR gates.
MS2 and turning it off
The MS2 hairpin has shown tendencies of not being equally easy turned off, depending on which side it is compared to the ligand (microRNA etc) it binds up with. So things like this will play in too. Which may also be why MS2 turned up later in the OR lab, instead before. Just as the Exclusion 4 lab scored better than Exclusion 1. Exclusion 4 had MS2 last, and Exclusion 1 had MS2 first in the RNA sequence.
Oligo strength content and landing spots
Things like the individual microRNA/oligo base content and strength will of cause also affect things. But I think there may even be patterns to how they prefer being placed.
I think we with time and more labs will be able to read preferences in relation to what type oligo needs to land where, depending both on lab type (OR, AND, XOR etc) and ligand personality.
Relation between size of switching area and number of switching states
Another thing that I was earlier wondering about, now seems to have found its answer. I was wondering about the amount of switching stems for puzzles where there were more states than 2. If stems would need to get longer or shorter. (Background post)
From what I see till now the switching area actually get bigger - with more stem area involved - which kind of makes sense. Since the RNA needs to fold in multiple states, it gets pulled in from multiple possible solves, meaning that longer switching areas are needed. However the stems still typically get broken up in shorter bits, so they don’t get much longer than in 2 state switches. Exceptions are for ligand bind up sections.
The new thing is that instead of a switching stem, there is more like a thing as a switching arm, like two stems with loop between them.
Now we have good scores for starters, I expect we will be able to reduce the switching area slightly for some of the labs - I expect even multi state puzzles to need as small a switching area as possible, despite the switching area will need to be bigger and not just be about two stems swapping strand partners. Other variations are possible also for the two state puzzles, just cutting my point out in cardboard.
1 base difference - 10% score difference
One thing I found particular interesting when I was reviewing my designs, was that in one case, just one base difference, resulted in a 10% score change.
No 1-1 loop between last oligo and RNA design, score 60%
A 1-1 loop between last oligo and RNA design, score 70%
Now I did make some other designs that also had 1-1 loops but didn’t score higher for that reason. Not all spots are equally optimal. I think a mismatch start to get beneficial when the stretch between oligo and RNA design gets rather long. Just as static stem favor not being too long, if they are to switch.
Also some of our earlier microRNA lab designs seemed to sometimes benefit from 1 or more mismatch between microRNA and RNA design.
I think a valid way to raise the score of several of the higher scoring designs for next round is to systematically mutate in 1-1 loops between oligo and RNA design, in the designs that do not already have a 1-1 loop. Some designs may even benefit from having 2 1-1 loops. At least when it comes to the longest oligo binding up. The shorter oligo may not have such a strong need for a 1-1 loop.
I think I found a way to weed out some of the bad designs before we submit them. My advice is to try avoid making designs that have more than 5 kcal difference between 2 and 3 state. Here is why.
There seems to be a pattern for the energy difference between state 2 and 3. So far the topscorers in the AND and OR lab have a difference between them of something like max 4-5 kcal. I can’t say much about XOR as we have so few high scorers there.
Whereas some of the lower scoring designs have a bigger kcal difference.
Although being in a good kcal difference range is no guarantee of hitting the high score. Just like entropy in our static labs. It had to be low, but having low entropy was no guarantee of making a winner.
Here is one of my AND lab design that scored 30% and has a big kcal difference between state 2 and 3.
Similarly I suspect but am not yet sure of, that having too similar kcal between state 2 and 3 isn’t beneficial either. But that I'm less certain off.
I expect there may be slight variance of kcal difference between state 2 and 3 in the other labs that we don’t yet have data for, but I count on this pattern being general for the other labs too.
Both state 2 and state 3 binds to each their oligo and as such will be similar in energy. However the oligos are different. Both in length and strength.
Also there is a trend for which of state 2 and 3 that has most negative kcal. Which is typically the state which has the short strong oligo attached. Now many labs can be solved with oligos in both orders, but I think a lot of labs will have preference towards a certain order of the oligos as I pointed out in the post above. Which means we in the future can say something about which state of 2 and 3, that should take preference and have the more negative kcal.
I have noticed that the two bases around the MS2, tend to be the same in the high scorers, for each type of logic gate labs. (XOR, AND and OR).
I think some of the neighboring labs with similar amount of MS2’s in on and off position may want something similar. Which isn’t too bad knowing.
Showing the full range of neighboring labs, in same order as above.
Examples of having same bases around MS2
OR lab, notice the U base before the MS2 and the A base after
Here I made the sequences line up with sequence search, since most of the top scorers contained same sequence pattern after MS2.
AND lab, U’s on either side of MS2
The OR lab needed the MS2 stable and on in 3 states and had an extra AU base pair extending the MS2
The AND lab needs to have the MS2 unstable and gone in 3 states, here the high scorers tend to have an UU mismatch around the MS2.
MicroRNA labs with single input versus Logic gate labs with double input
Looking back at earlier MS2 labs, also many of the microRNA lab high scorers have some specific likely base suspects around the MS2 for each lab.
In the 208a lab, the two bases around the MS2 were pyrimidines. Either as a UU mismatch or a UC mismatch.
Here my mod of Mat’s mod of JL, score 100%
The Sensor V3, variant 2 lab had an AC mismatch and sometimes a pyrimidine mismatch.
Salish mod of Mat mod of JL, score 100%
The turnoff lab Sensor V3, variant 1 had a GC pair. This lab we had a hard time solving.
Perspective on lab type and bases around the MS2
The microRNA labs don’t seem to follow trend I mentioned for the current logic gate labs, with the lab having most MS2 turned off, also having the bases around MS2 that would be most destabilizing and vice versa.
In the logic gates, the overall MS2 turnoff labs, having loosing up mismatches, where the overall MS2 turn on labs have a stabilizing base pair in front of the MS2
I think the microRNA labs with just one input and two states, versus 2 inputs and 4 states are far less pressured.
The riboswitch on a chip labs also have some trends for their bases just around the MS2. In the exclusion labs, the one of these bases are typically locked - due to the MS2 needing to be very close to the aptamer sequence.
It also seems to depend a bit on design types. In the Same State NG 2 lab, the PWKR design type that needs to be switching big time - almost full moving switch - except for the neck, has the UU mismatch around the MS2 - which I suspect is generally rather helpful for getting things. moving. JL’s gliding switch type has UC or AC mismatches whereas the Xeonanis type of the winners have what ever base that is embedded in the two static stems around it. But typically 2 G’s. There are several more variants. The Same State NG 1 and NG 3 shows clear trends bases around MS2 for either GA or AC, following the hidden hairpin in a loop pattern.
Jieux mod, score 96%
So generally it appears to be a bit more variable for some of the Riboswitch on a chip lab compared to how much similarity there is in the microRNA labs and especially for the logic gates labs.
I believe MS2’s next door neighbor bases are worth looking out for in future labs. I think the amount of MS2’s needed either turned on and turned off will affect the bases just around the MS2 a lot. But turnoff and turn on and in between lab types, matter, things will also get affected by the type of input the switch is getting. Like molecules, RNA oligos and the number of them as well. Even what switch element is used, I expect will affect things.
I have been drawing some expectations I have in relation to our new labs, based on the sorting of the labs below and trying to solve them. A few of the drawings are based on actual lab results (OR, AND and XOR). Similar I have gotten inspired by some of jandersonlee's mutants and some of my own designs for others of them.
Labs ordered after how many MS2’s are off, ranging to how many MS2’s are on
This has been my working hypothesis for this lab.
This overview drawing below is related to the one I shared in this post earlier and have re posted here. Notice that the weaker oligos come first in the left column and last in the right column.
I expect some reversal of oligos as possible and wanted for some of the labs. I think some labs will exclusively prefer one order of oligos over another (In the AND lab this seems to be the case already) and some labs being able to tolerate more different oligo orders. But I also expect that most labs will end up with a specific preference. A majority type over minority types.
Here are some more detailed drawings of individual labs. Here the lab drawings comes in the same order as the labs are ordered above. For some labs I have drawn several.
Of cause most of this are guesses and there are minority versions for some of the labs. All the details won’t be right. But I hope this can be helpful for designing and give ideas for breaking down the problem of how to achieve a turnoff of a specific MS2.
Based on lab results
NAND oligo exclusion type
Based on lab results
Based on lab results
Based on lab results
Reuse of structure between state
The more states you can let reuse a bit of the same structure or turnoff sequence, the better.
In 2 state switches, it has generally worked well to limit the number of actual switching bases, by making static stems and non moving parts. The more bases that have to switch in the first place - the bigger the chance that something will go wrong.
The same principle seems to also be in play for the 4 state logic gates, although in a little different manner. Here you don’t necessarily have to lock up a stem in all states, but just it being the same in some states seems to be beneficial.
Structure reuse in 2 states
Example with structure reuse in top scoring mod by JL in the AND lab, 88%
Here a whole arm is reused. This takes advantage of that the oligo binding site early in the design, needs to be gone in both state 1 and 3. No need doing that in two different ways.
This even happens twice in the same design.
Here it is state 1 and 2 that both needs to have the other oligo binding site packed away.
Structure reuse in 3 states
Example with reuse of structure in logic gate high scorer, 88% from the AND lab.
Here the recycled stem consist of two strands that are paired to hold MS2 for turnoff. This is the same in all the 3 states where MS2 needs to be gone.
Oligo complement and oligo like straight pair up
The logic gate gates puzzles don’t seem to like having bends between oligo touch up area and RNA design. Although one of my designs did somewhat okay with it, I think it is not optimal, at least for now. I see far more bends between oligo complement and oligo among the lower scoring designs.
Even just a 1-nt loop bulge can be trouble. Also the more bulges, the worse. The more bulges and bigger bends there are between oligo complement and oligo, the bigger the chance that the design will end up among the lower scoring designs.
I think that a slight bulge or loop bend, raises chances for the design to kick off the oligo. Also more than GU’s, 1-1 loops and more even spaced internal loops - although these works for oligo kick off too.
Turnoff labs versus turn on labs
The earlier microRNA labs with one RNA input, the turnon lab 208a liked straight pair ups between microRNA and complement also, while the turnoff labs would better tolerate some more variations like having the microRNA landing spots split between two distant regions in the RNA design. This pattern only turned up in a minority of the 208a winners, where it was the main pattern for the turnoff labs. But still no bending in the contact between the microRNA and the RNA design.
However the OR lab where I did get away with the bending bind to the oligo, is a 3 state out of 4 MS2 turnoff. (Image above) So my hypothesis for now is that in overall turnoff labs, a bending catch of the oligo might be a way to easier get rid of the oligo when needed. And that overall turnoff labs are less harmed by it.
Although it is probably bad news for my exclusion strategy I used for some of the A/B labs, where I used the presence of either switch elements to kick the other in a very direct manner by exploiting overlapping base regions between both oligos or between oligo and MS2. This typically also leads to a bulgy binding.
So while I’m not sure how useful bending contact between oligo and oligo complement will be for now - as it seems to be best to avoid it - I strongly suspect it will come in handy for another situation.
Later when we get more RNA inputs, some bending may be helpful as a mean to disrupt pairing. Like if we want one microRNA to bind less well than the others. Also if we should get some really strong based microRNA that would be far less likely to unzip when first bound and we need to kick it off.
Making the oligo complement fold with the other oligo complement when they both needs to be gone, is generally an easy solution for the 1 state in the logic gates lab. However it is not always optimal. I think it is sometimes more risky as it easily ends up locking the design up and leaving it less willing to switch. I think when to use it, depends on what kind of lab puzzle it is. If MS2 mostly needs to be turned on or off. And in particular in which state.Here is how my investigation of this started.
Breaking up longer stems
I was looking through the OR designs when Malcolm’s design OR Gate #56 caught my attention. Because I got curious about the design and its siblings, I pulled a sorting by, on this design.
I found it interesting that just two mutations could render such big difference. Although they do break up a too long static stem forming in the one state.
Notice that the long switching stem in state 1 (left), gets broken up into 2. Causing the score change from 40% to 69%. I have earlier mentioned that static stems can get too long and thus prevent the switching. So not too surprising.
Now you may wonder why there then seemingly is an equally long bound stem in the OR lab high scorer. I shall explain below.
The oligos can better handle being paired fully up to their oligo complement than the designs can handle having long stems without GUs relaxing them, 1-1 mismatches or internal loop helping the stems break free and switch.
Magic wand or bunny ears
I have been wondering about why the designs that had one half of the sequence pair up with the other half had so different kcal to those long stemmed designs we had in the past static labs. However there was a rather simple explanation of that one. Loops...
Kcal just a very rough average, not to be taken too literally
I simply took an old long stemmed winning lab design broke some of the base pairs in the stem and put in internal loops and got in a similar kcal range as for our state 1 logic gate puzzles, with oligo complements folding with each other.
Static versus switch lab - GU’s, mismatches and internal loops
When we a long time ago solved static labs with real long stems, GU's turned out to be welcome and almost needed in longer stems - to keep them happy and static. Also mismatches worked well for that, just as Brourd and Nando thought. Also designs that kept a low GC count also easier got away with avoiding GU and mismatches. But else GU's or mismatches were overall omniscient in the winners in this type of lab design.
Background post: Designs that crave GU
Back then I also noticed that GU's turned up in switch labs in excess. Now 1-1 mismatches and bigger inner loops in between longer sections of switching stems to a higher degree plays a role, while the GU's are still regularly necessary. If there are too few GU's, too few mismatches or too small loops, score doesn't get good.
I find it kind of fascinating that both long stems in static labs can be stabilized by loosing up base pairs like GU or even mismatches. Whereas the same base combos can be aiding a switch getting switchy and moving. I think the point of difference is the length of the stem these turn up in. In a static lab GU’s would turn up in real long stems, as in switch stems they would generally turn up in shorter stems that are supposed to switch.
Structures in the logic gate labs
I have been really annoyed with the longer stems forming in some states - state 1 in particular. I called this for the magic wand shape as opposed to the dumbbell or bunny ears shapes, that I decided I would rather see. I ended changing my mind a bit.
Example of first half of the RNA sequence pairing with the second half.
Example of bad AND design, score 67%
Thinking about it, this long design probably have too small loops, though it has a good amount of GU’s. (The designs also seems to go grumpy if they have 3-4 GU’s really close). Bigger internal loops - though they can probably get too big - seems to help get things opening up.
Here is an successful design that avoids to have half design pair with other half in state 1. Shape bunny ears.
JL’s top scoring mod from the AND lab. (Score 88%)
Notice it has far fewer switching base pairs. Plus it doesn’t have really long switching stems. I think it is easier to get a switch if the oligo complements each can be made fold with themselves for oligo landing site turnoff.
That way there are fewer switching stems in line that needs to be broken up. The fewer switching pairs, the easier switching. Despite it seems the extra RNA inputs make the logic gates labs more tolerant to bigger switching area, I think there are limits.
The reason why I have been confused with the really long stretch designs are the following. I been seeing this pattern turn up in disproportionate amount among the lower scoring designs in general. However it clearly works in the OR lab. As I did see a 13 base pair long stem pass in the OR top scorer. However it also had 4 GU’s. :)
JL’s Brourd mod, score 88%
While this one pass - likely due to its GU’s - what I do know is that I really would have liked there to be ban on stem length outside of oligo binding area, over a certain length. :) I see way too many long stem stretches. And they are hurting the switching ability. The longer a switching stem, the bigger chance that it is not so switchy at all.
The winners in the Cooperative lab basically show the same pattern (Magic wand). Two MS2 oligos pairing up for mutual turnoff. Just directly and not indirectly as with the oligo complement landing spots.
When to go magic wand versus bunny ears
I think I figured when magic wand structure happens or at least is acceptable.
I believe in bunny ears and dumbbell shape for state 1 if MS2 should be shut off. However if the MS2 needs to be on and is in between the oligos, magic wand shape becomes more needed or acceptable. This at least explains which labs the long wand shape turns up in. These longer paired stretches tends to turn up if the MS2 only needs to be off in one state. (Though we don’t have data for most of the labs yet)
AND logic gate - MS2 Turnoff Sequence Recycling
When looking at the AND gate high scorers in the logic gate lab, what I find particularly interesting is the turnoff sequence after the MS2 or rather in prolongation of it - UUGGG. It is the exact same that was successful in the Same State 2 and Same State NG 2 lab, in the Xeonanis type design that started over a ViennaUTC design.
Red box = MS2, orange box = MS2 turnoff
JL mod, score 88%, color highlight match with the boxes above.
Example from the Same State NG 2. MS2 and FMN turnoff sequence UUGGG after the MS2.
Mod by Mat, score 100%
I simply love that it seems that some elements like MS2 surroundings are sometimes transferable between switch labs. Admitted that this time I put it there. :)
The GGGUU works with same purpose as in the original lab Same State 2/NG 2 that I took it from. The GGG’s are targeting the MS2 C’s for turnoff. In the Same State lab there is just a static stem in between MS2 and turnoff sequence. Which there isn’t in the AND lab. It does something else. It seems to create a small switching stem consisting of magnet segments.
Notice the tri loop too. I have used tri loops several times for switching stems with the intention of using the fact that the tri loop should be easier to break than the more stable tetraloop - even if unboosted. Now this GGGUU turnoff only works for one of the states, the puzzle has 2 more MS2 turnoff sequence.
This small switching stem created between the MS2 turnoff sequence and bits of the MS2 that seems to pair in state 3, brings me to a new pattern I see emerge. Really it is actually an old pattern. More on this to follow.
MS2 ON structure
A structure like this (Orange box) with a MS2 followed by internal loop followed by switching stem. This structural pattern regularly shows up in one of the states where MS2 turned on.
Example from one of the AND lab high scorers, Score 88%
I am having a clear case of deja vu. :)
This looks suspiciously much like the RNA blueprint pattern that I pointed out for the riboswitch on a chip/NG designs - just without the static stem in the switching area.
Drawing from the above mentioned post, cut out and drawn upon.
Ignore the static stem in the switching area in the Same State NG 2 lab and we basically have the same situation in the logic gates, in one or more of the states where MS2 needs to be on.
I have an older drawing of the patterns in the Same State designs. Here particularly notice the Same State NG 2 type. Here is the same magnet segments in the aptamer gate and they can turn either way. Though one type is favored. (left one of the two - Xeonanis)
Without the static stems
Specifically in some of the MS2's turned on states, I basically see a internal loop between the MS2 and then a switching stem - often consisting of magnet segments. Just like in the Same State labs and the aptamer gate with its magnet segments. And just as in the Same State lab, the magnet segments could be reversed. (see drawing above)
Here is a drawing I shared earlier when I was thinking about switch bubbles between the switching elements.
The microRNA lab contained an internal loop, where the Riboswitches on a chip/NG labs contained a multiloop in their switch bubbles.
However this switching bubble has seemed to go more extinct, the more RNA inputs and states there are. It seems like, the more RNA inputs, the more opened up the RNA design seems to be. Or rather some of the labs. The XOR wasn’t though. And the OR lab certainly got bound up in State 1, with first half of the sequence pairing with the second half.
Switch bubble and MS2 turn on
However it seems the presence of a turned on MS2 that starts to call for a switch bubble. Not in all states with MS2 on, but in some. There may be something about this switch bubble that helps MS2 gets turned on.
This switch bubble is regularly followed by a stem made of magnet segments. Although less than magnet segments for the switching stem will also regularly be enough.
Switching stem with magnet elements
For this lab round I had attempted to put in both magnet switching strands next to the MS2, far away from the MS2, as prolongment of oligos, but also use something as a switching magnet stem with magnet strand elements in between one or more of the switch elements.
Ok. What do I mean with a switching magnet stem? I mean a stem that is involved in switching and consists of magnet segments. It could be mostly G’s in one strand and mostly C’s (some U’s allowed) at the partnering side.
Here is the first case I have seen of a full blown switching magnet stem.
This switching magnet stem pattern even turns up in a big numbers of the XOR labs that are yet without lab results. More on this to come.
Example from NOR and end positioned MS2.
Sometimes the internal loop next to the MS2, will also contain magnet segments too at one side. Like below.
I think what strategy is most effective to use when turning off MS2, will to some degree depend on in how many states MS2 needs to be turned off in. Also to some degree which states. State 1 and 4 sometimes behave somewhat different, whereas state 2 and 3 can often can share strategy of approach. Here is what patterns I see for now for our 4 state logic gates. Below I will post examples of each type I mention.
Tail magnet/Switching magnet stem
Short range MS2 turn off sequence - strand magnet/switching magnet stem
Far range MS2 turnoff sequence - tail magnet
Short range double strength turnoff - Mini MS2
Oligo complement for turnoff
Tail magnets and switching magnet stems
I got inspired by pattern I saw start turn up in the XOR labs. I first noticed this pattern back last year before we had any results back from the XOR labs.
Here was what I wrote:
Here is a pattern I have seen grow in presence in our XOR puzzles in general.
I traced it to Brourd for the microRNA lab Sensor 208a. The pattern typically occurs in either end of the RNA sequence. And is regularly used to shut down the MS2 when it isn’t needed.
However thinking about it, I think it is kind of an universal pattern for use all over the puzzle - in multiple state switches. As it helps bury a sequence when it isn’t needed, but also be ready when it is.
The pattern in this puzzle is really kind of a mixture between being a magnet switching stem and tail magnet.
This exact pattern in the microRNA design above, was a weaker minority pattern among the microRNA 208a designs (Single RNA input RNA labs) - not yet capable of showing its full strength. Only a minority of the winners ended up carrying it.
The logic gate puzzles mostly don’t contain this exact pattern, but rather some strand tail magnets for long range turnoff or some switching magnet stems which can be placed at many different spots.
As I expected this tail magnet strategy has turned out much stronger in the 2 RNA input logic gate labs.
Short range MS2 turn off sequence - strand magnet/switching magnet stem
If MS2 is to be turned off in most states, then short range turn off or sometimes turnoff sequences on both sides of MS2 are generally most helpful. (Next door neighbour turnoff. Can be by single strand or more)
Such nearby MS2 turnoffs are often addable. Also nearby magnet segments in the oligo complements can be engaged as MS2 turnoffs.
On why adding starts to be needed. When MS2 needs to be of in several states, it will start to need to have turnoff sequences at both sides. Except when state 1 and state 2 or 3, can share turnoff structure.
Here are some short range MS2 turnoffs already seen in earlier MS2 labs.
Far range MS2 turnoff sequence - tail magnet
If MS2 only needs to be turned off in one state - a long distance turnoff will often do. Like a tail magnet strand or oligo complement prolonged with a magnet segment. This will particularly work well in state 2 or 3 at the side where the oligo complement needs to be gone too. Then the oligo complement just have to fold up with itself - it can also be helped by a little extra added complementarity at either end. Then magnet C or G turnoff sequence, depending on if you target the MS2 C’s or G’s, just needs to be placed at the tail end.
Tail magnet turnoff can’t be practiced in state 4 where both oligos are to be bound, since this strategy requires that the oligo complement folds with fold up with itself, to bring the tail magnet in close to the MS2 for turnoff.)
Abbreviated MS2 for MS2 turnoff - short range
When half of the MS2’s needs to be of and the other half needs to be on, it seems like using a mini MS2 turnoff can come in handy. It turned up in Nando’s XOR top scorer. I first saw this Abbreviated MS2 in the Exclusion 2 lab in Perushevs design.
Far range double strength turnoff - Mini MS2
The mini MS2 turnoff can probably also be done long distance. I just have to see it in play yet.
Function of Switching Magnet Stems
So far I see switching magnet stems may serve two purposes. Both securing a strong hook up as individual strands when the switching stem is broken, but also provide a helping hand to get the MS2 a quiet place to get turned on. Which brings me to something new.
One big difference in the logic gate labs compared to earlier MS2 labs, is that MS2 seems to become a notch weaker, when around 2 other microRNA input, where it dominated the dance in the single RNA and molecule input labs. In the 2 state Riboswitch on a chip in the Exclusion labs (turnoff types) - MS2 was generally only way too happy to fold with itself - the problem was rather getting it turned off.
But when MS2 is playing with longer microRNA inputs, things seems to change. I think this is also why we see it is now allowed placing MS2 at ends of the RNA sequence as in the OR lab top scorers, with little to turn it off.
Where we in past Riboswitch on a chip had trouble getting MS2 turned off if it was not force held by both ends - or as in the Exclusion NG 1, Exclusion NG 3, Exclusion 1 and 4, being turned off by a longer aptamer gate, compared to what was found in the easier labs. And in the Exclusion 2/NG 2 case, if not holding most of the MS2 tight with an abbreviated MS2 for MS2 turnoff.
I think switching magnet stems is playing the same role that the aptamer gate and FMN molecule does in the Same State 2/NG 2 lab. Helps provide switch bubble and a switching magnet stem (the aptamer gate) for MS2 and FMN to get close to each other. The magnet stems in the switching stems, can both be used for MS2 turnoff and for turnoff of oligo complements
Oligo complement magnet prolongation - short and long range
Magnet segments don’t need to be next or near to the MS2, they can also be placed as prolongation at either or both ends of the oligo complements.
Prolongation with magnet segment of oligo complement has potential to be both short and long range MS2 turnoff. But will also often be used to shut the oligo complement itself off.
Example from JL’s high scoring mod (score 88%) of my design, prolongation shutting off MS2.
Even magnet segments inside the oligo complements can regularly be used for MS2 turnoff. Especially if they are at ends of the sequence.
Here's a sample. Note in particular the nifty new summary graphic Johan has added in the upper right hand corner. The Target rectangle shows the desired level of KDs (blue representing high and and yellow low), while the Results rectangle shows what was actually measured.
Omei, I noticed a real neat detail in that spreadsheet. I can see the graphs directly, without having to open anything extra. This is simply awesome! :)
I have earlier described the process of designing for static labs. I have tried make a demonstration of picking and mixing favorable mutations in the XOR lab, when making mods of previous round lab designs. Plus a bit of the process before.
As for the latest XOR round I have been especially interested in mutations between oligo and oligo complement as I saw that just one single base mutation, leaving a 1-1 loop between an oligo and its complement, having a quite profound effect in score change.
In this analysis I have made use of the fact that many of the high scoring designs are sibling designs - meaning they are mods of the same design or from a design series.
I began looking for mutations that raised score between designs in the OR lab, plus including a mutation that wasn’t in the top scoring design of the round, with intention of adding and mixing them. I pulled a sort on a design I was interested in.
OR Gate #59 and OR Gate #58 has one base mutation and a score change of +6%.
This mutation is not in the jandersonlee mod high scorer. So making a mod adding the mutation in.
jandersonlee’s Brourd mod, score 88%,
I noticed that two other design had a 1 base mutation and a score difference on 3%. This mutation is also not in the high scorer. So I make a mod.
Score difference 14%
Mutation is already in top scorer. Ignoring.
Pooling 2 or more mutations
I decided that it might be beneficial to add two individual good mutation together in a design, if that combo has not already been made.
Adding G30 (+6% score potential) + G79 (+3% score potential) mutation in the top scorer.
Mat - OR Logic Gate - JL Mod - D4, score difference 2%
Pooling U84 + G30 and U84 + G79 and U84
3 base mutation
Pooling U84 + G30 + G79 and added in the high scorer
Pooling good mutations will not always result in higher scores. As the mutations may counteract each other. It also matters where they are placed in the design.
Some mutations seems interrelated. Like GUGGG or GAGGG at base 15-19 seems interrelated to if there should be 2 or 3 G’s at base 46-48,49.
Mat sent me this image, which I think goes to illustrate well.
As he said:
Mutate at 15 G to U and 51 U to G is the only common change in those 6 designs, but it does show how two mutations can affect the design/sequence/score.
Another approach is to isolate the regions most likely to change. And work through the changing base sections 1 by one.
Here I pulled a sorting by the top scoring design.
For base 1, it is quite obvious that the G mutation is far better than the A.
NB, here you need to keep an eye on the scores too. The designs are sorted by similarity to the top design, and not by score.
So for many bases, one base is the more optimal over the others. Of cause this do not say anything about those bases not tried out yet. But when there is a large consensus of base between higher scoring designs, I think it is less likely to hit a beneficial interchange mutation for that spot.
Or rather, I think pooling potentially beneficial mutations is a strong (in fewer mutations) road to higher scores - when doing rational designing based on lab data. But that single base mutation beside that, is a good supplement. Both will do well, but I’m after those specific base mutations that I strongly suspect will make a change.
Different family clusters
While many of our MS2 switch labs have had one main solve type that hits strongly though, not all labs have just one majority structural type among its winners. Some labs also have minority type designs too. This has effect on how one transfers beneficial mutations between one design to the other.
Illustration of different family clusters. Top one is the jl-olg-1.06 Brourd mod design cluster, the bottom one is the Eli algo 1.32 design cluster. They have different orders of the oligos also in relation to MS2. While Oligo 1 and MS2 are somewhat lining up between the different family cluster types, Oligo 2 isn’t. These design types are absolutely not the same.
Sorted by the OR Gate #56 design, but with an added minimum score. Majority cluster on top, minority on bottom.
That is why we can't mix beneficial mutations between all top scoring designs because if the designs belongs to different cluster families where oligo order differs, they are too different. Then what is good in one will most likely not be good for the other.
Basically I think 1 and 2 and 3 base score raising mutations are potentially interesting to try out and put onto a top scoring design that is related, but with a certain mutation distance. Perhaps max 10-15 mutations difference to the smaller mutation distance designs where the beneficial mutations come from.
Mat and I have been discussing. The following thoughts are the results. Thanks to Omei for discussion too.
I had noticed that adding in a 1-1 loop mismatch between oligo and oligo complement regularly seemed to be beneficial.
However the discussion with Mat, made me realize that these 1-1 loops may pop up somewhere specific. The images he sent me on an unrelated discussion, even made it stand out very clear.
Similar my discussion with Omei strengthened my feel that there might be certain positions that were better to put in 1-1 loop mismatches.
Where do mismatches between oligo and oligo complement occur?
A good deal of the top scoring designs in the AND lab, avoided using the strongest possible pairing between oligo and oligo complement - at both oligo sites. The mismatches turned up at specific spots. Namely somewhere in the middle. Actually in the AND design, at exactly the same spot.
Score 88%, jandersonlee mod
Now many of these designs are siblings - mods of the same design/s. And as such the likeness is natural. I however do find it interesting that the majority of the high scorers in the AND lab make it to the top containing this pattern.
Discussion with Mat
Mat has been working with idea - elimination of unwanted sequences - like sequences that will generally lead to fail. With intent to avoid these for the future. And idea I’m sympathetic to.
There are sequences that will be generally bad for certain size puzzle and type of switches. Just as there was for static labs. He had several sequence candidates, however one of them made the conversation go in a new direction. Mat sent me the big UUUUU picture below, which started the following discussion.
Sequence elimination discussion
Mat: I was trying to see how some of my single mutations lowered the score
Eli: That’s the reverse way of looking at it. Looking at which mutations that lower score, is also telling something valuable. What to potentially avoid, later. Although one can’t rule out base mutations since it’s the combo’s of them that matters. A mutation may be bad if it doesn’t have the right company.
Mat: Then I thought i started to see other patterns in lower scoring designs, so I start to do some search of sequences to see if some patterns were only found in lower scoring design
Eli: Usually UUUUU is a classic to avoid. But I have started using again because since we have gotten barcodes in, they haven't been as critical and I saw them do good in a few cases.
Mat: yes I seen one of your uuuuu
Image that Mat sent: (UUUUUU’s being bad in OR lab)
Image by Eli: (UUUUUU’s being beneficial in AND lab and at least not harmful)
Eli: I have particularly used them for oligo complementary site. Else I think long UUUUUU repeats are bad. But I think they are sometimes beneficial there as while one want a general strong match, between oligo and RNA design, making it too strong is sometimes problematic. And the easiest change over for a C base is an U. [Correction here an A was replaced with an U and making a mismatch, leaving the middle of the oligo complement broken by a mismatch, I wrongly remembered another case where a C was typically replaced with an U and also making a mismatch.]
Mat: yes I don't think all of the pattern/sequence I seen/screenshot were bad all the time
Additional info on barcodes
As Nando recently explained:
"Barcodes are modeled in the puzzles only for SHAPE experiments, because there is a potential for interference. I think Johan mentioned also using barcodes for his microarrays, but they aren't transcribed, so we don't need to worry about them."
Displaying oligo binding site
Mat sent me some spreadsheets as he was playing with an idea of how to get a visual of the placement of the oligo complements. He showed me.
I have shot images of the sheets as I loved how visual the 1-1 loop mismatch site stood out in them across the highest scoring designs.
Notice the U in the middle of the red colored oligo complement A. (The designs are sorted after score, top down.)
For Oligo A
I added some blue to highlight the middle U mismatch.
For oligo B
With A mismatch in oligo B highlighted
Mat’s vision behind the spreadsheets
What Mat wishes to illustrate with his spreadsheets, is for is a way to make the oligo complements and the MS2 to stand out in the spreadsheet data for analysis. I second him. This would be a valuable feature. Both in spreadsheet and data browser.
Here it is quite obvious that the oligos have a favored spot and order. This is what I have been aiming to tell by drawing structure images. But here it just stands out naturally.
Similar here the MS2 position in the designs is highlighted.
The nature of a microRNA
I have been thinking about microRNAs/oligos as kind of weak hairpin stems. Imagine the oligo folding and making a weak hairpin with itself. The middle of it being the hairpin loop. Its ends should be able to fold weakly with themselves, but also they should not be too happy about folding with themselves.
Now imagine the same scenario for the oligo complement. It being able to make a hairpin stem loop with itself. Which is actually what we regularly need to do when we are packing the oligo complement away in one state or more states.
What does the middle mismatch do?
Hydrogen bonds are rather weak in themselves - but the more of them added together - the stronger the bind.
What happens in a stems in switching puzzles? When a stem is either real long or solved with strong bases - you ain’t got no switching! Long stable stem regions makes for bad switches.
The 1-1 loop mismatch is like giving the design an elbow joint. And the oligo when bound up to the oligo complement, becomes an arm. Instead of a long and strongly bound and less flexible stem.
What I think happens is that having an oligo with a somewhat weak middle or not fully binding up at the middle, gives the design extra flexibility to shift between states. If the middle bit of the oligo complement isn’t as hard bound as the two outer stretches.
Just like folding paper, you are giving the puzzle an extra spot for folding.
As a positive side effect I have noticed while designing, that often a mismatch or weak bind like GU, placed somewhere in the middle will actually allow you to make a longer stretch of the oligo pair up with the RNA design, than you can get done without. Having a good deal of the oligo actually binding up with the design, improves your chances of catching it.
So with a break in the middle of the oligo complement that should bind with the oligo, I think basically there are more places to for starting zippering on for the oligo and similar more places to start zippering off.
It seems that middle placed mismatch helps the oligo drop of the complement.
Still while each of the oligo complements in the design being capable of base pairing like 8-10 base pairs with the RNA input. I simply think this speeds up both the attaching and
detaching process. The 1-1 mismatch has already played that role in the microRNA labs earlier.
Or put opposite, I think that having all bases in the oligo bind up, instead of having two sections bind up, slows down the switching process and makes the switch less effective.
Perspective on mismatch between oligo and oligo complement
I think there generally is benefit from having the oligo complement pair well up, but with a break somewhere in the middle, as soon the complement gets of a certain length. Most 10 base sequence likely won’t need them, while more 20 base sequences will.
Not all logic gate labs may need mismatches between the oligo and oligo complements or not each of them. Not all will need them at the same spot. It will depend on what you need the switch to do. But I think this middle mismatch pattern is worth looking out for. For when and where to use it.
In some cases it may also be useful to avoid mismatches on purpose, should we for some reason need to slow down the speed of one state in comparison with others.
A mismatch can also turn up at other spots for other particular reasons which I shall get into later.
Something Mat said a couple of months back got me inspired. I have been thinking about it since.
Mat: "I see you wrote about some of my riskier strategies when using numbers of GU`s in a sequence/design. You were talking about G magnet segments and C magnet segments in the forum. I was putting in weak point in the sequence, due to them be strong."
I find what Mat said particularly interesting as it highlights the contrast between both strategies - strong magnet segments mixed with weaker and breakable areas. Something that are both signatures of switches.
What he said reminded me of another thing I see in the switches and in particularly the logic gates. The whole sequence is practically a shift between G magnet segments and C magnet segments taking turns - or pyrimidine and purine sections. They go through it all like a red thread, with some other bases in between the magnet segments.
Now this is a very simplified drawing of the switch RNA sequence and its magnet segments. The magnet segments will not be exactly placed at every second spot. For a number of reasons which I shall come into later.
Here the magnet segments take turn, except before the MS2.
OR lab, score 85%
Also I should mention that some of the magnet islands are of cause decided by the sequence of the RNA input/s and as such will change with other microRNA sequence. And where they will be, will be depending on where the inputs are placed in the puzzle too.
Others magnet islands are to make the MS2 or inputs turn off. Magnets around the MS2 often shows reversal in magnet pattern, which breaks the green red green pattern.
I expect the magnet segments around the MS2 to show some more pattern to them both for position also for RNA structure, than the microRNA input landing spots, as those will need to change with the microRNA base content and strength.
Also I will expect us to be able to reuse a lot of these, also between labs. With the same type of lab being able to reuse the same type of MS3 turnoff. Just as I was able to move a working MS2 turnoff from a winning jandersonlee design mod in the microRNA 208a lab. I took it from a single input lab and made it work in a new setting in both the A/B single input labs, Sensor A MS2 ON and Sensor B MS2 ON.
The magnet segments are what make things switch. They are fundamental to switching. Just as they were detrimental when too many in static designs or placed at positions where they could break loose and mispair with other complementary magnet segments.
In some of our early cloud lab switches we did have lots of magnet segments too. There we just seemed to have them in unlucky proportions, with too much pyrimidine ones over purine ones.
Reducing the amount of possible pairings between magnet segments
I did a fictive drawing of a RNA sequence with magnet segments shifted every second time. What I wish to illustrate, is how many possible pairings there already are with just a short RNA sequence and 8 magnet segments. Our current logic gate top scorers typically have like 6-8 magnet segments.
16 total possible pairings
Now just because each segment has 4 options of pairing, it doesn’t mean that each pairing is equally likely. A close by pairing - nearest neighbor strand pairing is far the most likely. Also if the magnet segment is in the early end of the sequence, and it in one state gets brought close to the matching magnet at opposite end that is a pairing that also easily happens, despite long distance in sequence. I did some drawings with some rough guess on. The thickness of the blue line indicate how likely I find the pairing.
How to make fewer potential pairing regions with the same amount of magnets?
One thing worth noticing is that the amount of pyrimidine and purine magnet segments when looking at logic gate designs. In the example below there are 5 purine (red) magnets against 3 pyrimidine ones. Just this small detail means there is one less total pairing option compared to had the amount been equal, (3x5=15) whereas the first drawing I showed of possible pairings with 4 red and 4 green magnet stretches, had 16 total possible pairings.
AND gate, score 88%
By swapping around some of the magnet segments, some of the pairing ups gets far less likely to happen. Notice the two red magnet segments after the MS2.
Two examples from the AND gate lab
Some designs carry the opposite pattern with more of the green magnets than red. Nando’s XOR solve carries 6 green magnets against 3 red. Which is a 6X3 of total of 18 possible pairings. This is still a reduction in comparison to if it had been 5x5 or 5x4 - accounting for that this design has more magnet segments and thus more options for potential pairing. As the number of magnet segments go up, it get more important to find a way for simplifying.
XOR gate, score 78%
Now for the AND labs that gave a near winning design, a 15 possible pairings are not much less than 16 possible pairings for the same amount of magnet segments when the same amount of each.
However this move by mixing the magnet segments so they are not red green red, or green red green etc in order, still achieves something that is way more worth than this seemingly minor reduction may lead to think - for another reason.
Magnet repellent sections
The red islands at base 30 and 40 will repel as they are same kind. Similar the ones at base 45 and 55. This means that by just by a few repeat magnet islands of the same kind, one can ensure that specific regions of the RNA design will be less tempted to do the otherwise obvious and often seen nearest neighbor strand pairing. Repelling magnets regularly happen around the MS2.
AND gate, Score 88%
State 1 and state 3 has sections that behave in concert. Marked above the line of each state.
Similar this design while actually having equal amount of green and red magnet stretches = while having 16 potential possible pairings, will reap the benefit of repelling regions due to the magnets being mixed and some regions repelling each other just by not having a repeated magnet segment patterns with one every second one being one of each. Not each of these 16 potential possible pairings of the magnet segments are equally likely.
OR gate, score 88%
Magnet segments behaving as a team
One can get two, three sometimes even 4 magnet segments to behave in concert and behave as they were one section - one strand - pairing up with other segments of magnet strands also moving in concert. Although often with room for loops in between them to avoid too strong pairing.
Thus one can secure a far more precise pairing up between sections, compared to if each magnet segment only behaved on their own. This strongly reduce the amount of interacting parts. Bringing the number of potential pairing regions down.
The highest scoring design in XOR by Nando, is practically 4 sections that behaves as each one strand or half of an arm.
All the top scoring designs in the logic gates solves have some kind of concerted folding of stems. Sometimes even with a reuse of these concerted foldings between states.
Nando’s XOR above and jl’s AND gate solve - concerted folds highlighted with gold stars
Logic gates as music beats
I have been telling Omei that I thought of the logic gate RNA’s being split in 4 section.
Basically when thinking about it, the MS2 and each the two inputs fills about a 1/4 of the RNA sequence length, each. These together take up a 3/4 of the puzzle. Then there is a 1/4 left that can be split in two or kept as a whole as in Nando’s solve above. This all reminds me of music sheets. Really Nando’s XOR solve is good old fashion rock. 4/4 beats all the way. :)
I have also been talking about each of the inputs as kind of being unwilling hairpins. Each being able to do a weak fold with it selves. So really some logic gate puzzles seems to come out with their own rhythm and 1/8 regions mixed in at different spots. The MS2 by default fills a 1/8 role too as it has to pair with it selves.
What is the point with fewer possible potential pairing regions?
The fewer potential options for pairings there are, the more likely, the ones you want happening are really going to happen. While good placements also matter, the former is not to be ignored.
Some magnet segments will acts as repellents of certain sections of the puzzle pairing up - by being placed so they will be unlikely to interact, other magnet segments will act as facilitators - by being placed so they will be more likely to interact.
Magnet segments are basically one more way of controlling what pairs with what, just like in static stems can be used as either fold facilitators or preventers. Both these strategies - magnet placement and static stems - can be played together and they will help us control what happens when and where we want it. Which is pretty beautiful. :)
Related post: Global fold change and switch simplicity