Round 97 Riboswitch Lab Discussion

  • 3
  • Article
  • Updated 4 years ago
It's getting almost impossible to find things in the original switch lab post, so I suggest we try a new conversation for each lab round.
Photo of Omei Turnbull

Omei Turnbull, Player Developer

  • 1008 Posts
  • 324 Reply Likes

Posted 4 years ago

  • 3
Photo of Omei Turnbull

Omei Turnbull, Player Developer

  • 1008 Posts
  • 324 Reply Likes
Much thanks to Eli and Meechl for getting a jump on publishing fusion tables and spreadsheets with additional fields.  And of course, thanks to Johan for the original data!

My first observation is that the dependency of cluster size on the percentage of adenine bases seems to have returned, after being pretty much absent in Round 96.  Here are fusion table graphs for the two rounds.

Johan, does that make sense given the process changes you made in round 97?
Photo of johana

johana, Researcher

  • 96 Posts
  • 45 Reply Likes
I think that the dependence will not go away. The new barcoding process ensures sequence accuracy and a minimum amount of clusters but it does not increase the number of clusters beyond that.
Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Hi Omei!

You got a good point. Adding a few links to the most relevant places for here for a starter.

Labs involved

miRNA Switch Lab - Round 3
MS2 Riboswitches On Chip - Round 4

Link to lab conversation related to round 97

Switch Scores for EteRNA Switch Puzzles
Johan’s microRNA Lab

Lab Round overview + links to spreadsheets

Spreadsheets for round 97

We should probably leave a link in the original Switch lab discussion post.

Loose thought: It would be good if we had a column in a WIKI table somewhere linking in forum pages with analysis related to those lab rounds.
Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Exclusion 2 - round 4 continued

Last round I proclaimed that I thought that Perushev’s design had a winner hidden it and that it held an interesting pattern too, because it reduced the MS2 gate by aiming for the opposite stretch to usual in the FMN aptamer. I have put up some analysis of that bit, so for background check here: The curious case of the missing MS2 gate

Now I want to look a bit closer on the difference between the original design by Perushev and some of its later siblings.

I have highlighted the only bases that were different between my mod and Perushev’s original design.

ChP 11-04-2015 #5, score 88%, cluster count 64, fold change 9.31

Eli x2 4.005, score 97%, cluster count 29, fold change 20.5

There were two base differences between our designs. I broke a strong base pair by changing base 18G to 18A and thus opened up a bigger internal loop. I introduced a weakness too in the aptamer turnoff sequence, to make things break easier by adding a G at base 36 - resulting in a GU in the switching area.

Also notice that I created a strong unpaired dangle of mainly G’s in the internal loop. Since it is in state 1, it has a chance of pushing the switch in the right direction towards switching to state 2.

These dangle G’s targets the MS2 C’s, just like in Perushev’s. But my added G’s not only creates a GU in state 1 but state 2 too.

I got a lower fold change and score in the one of my other designs where I didn’t put in the G base at 36 that created GU’s in both states but only changed the 18G base. So the creation of a GU in the switching area seemed to matter.

I got a slightly lower score and fold change for my design where I didn’t do the change that made a GU in both states.

Score 94%, cluster count 50, fold change 16.79

One design did get a better fold change than my winner, so despite my higher score, this design is probably the better one. As I think fold change is basically describing how well the design is switching. Fold change can not stand alone, good score matter too.

Mats mod of my mod of Perushev were 2 bases different from the original. One added U base in the static hairpin loop, which is probably not making a big difference. Whereas base 41U on the other hand removed the last bit of MS2 gate.

Mat - Exclusion 2 - Eli Mod - D4, 96%, cluster count 23, fold change 27.92

I find it very interesting that the one design that reduces the MS2 gate to non existent, is the one getting the better fold change. This fits well with the results of my experiment in Exclusion 2 and Exclusion 3 where I attempted making lab designs without MS2 gates, by following Perushev’s design. And it being well possible making winners in both labs, by altering strategy and aiming for the opposite to usual FMN stretch. Basically I think the MS2 gate that is often forming in front of the MS2 in state 1 in Exclusion labs, are slowing down the switch. True, making a MS2 gate is often the easiest and closest option of turning both MS2 and the aptamer off, by have a turnoff sequence on the opposite side of the MS2, to the aptamer sequence that is next to the MS2. And it has worked for making winners, but really I think it halts the switching.

I have a design where I removed the strong G’s from the bottom of the internal loop. It gets lousy score and crappy fold change. So apparently the direct targeting of the MS2 C’s are rather important, plus I consider the single base stretch flashing of strong G’s as a part of a switch enhancing mechanism. I think it is not just pyrimidine stretches that gets things going. Anything with a bunch of G's or C's in them and in single base regions are strong magnets waiting for a change of placement. I see more and more loop regions flashing G’s. Not just pyrimidine dangles.

Score 65%, cluster count 27, fold change 1.48

Ok, actually AndrewKae did manage to remove main part of the dangling G’s in the internal loop. And it still got a decent score. So I was probably too quick on the other assumption. He did some changes that enhanced the affinity of the end of the MS2 to the turnoff sequence, over targeting the C’s in the MS2 as the original design.

Score 92%, cluster count 35, fold change 18.9

I made one design where I only remove the base that was going to be MS2 gate. And just that base change is enough to improve fold change even if not changing score.

Score 88%, cluster count 23, fold change 14.13 (compared to 9.31 in the original)

I find it very interesting that in two designs, removing exactly that base (41) that makes a miniature MS2 gate, is enough to improve fold change a good deal.
Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Exclusion 1 - Round 4  

As I am a fan of knotting the tails of the RNA - yes, nature do make small switches that switches in the open end (Example below)  - I made a series of solves in the Exclusion 1 lab, that attempted to make static stem out of the two ends of the RNA, despite knowing they would probably be too short.

I did managed to score 78% with one design and 9th highest score in that lab, counting only designs with 20+ cluster counts. But only Vienna says that the tail section folded, neither Vienna2 and NUPACK said it folded, which makes it likely that it probably didn’t.

So as I suspected, making tails pair, only pays if they are long enough to actually stay together.

Also despite I think bigger designs benefit from tucking away the switching inside themselves, it doesn’t help if the tails are too short to add stability. This lab wants dangling ends. :)

The high scorers in Exclusion 1

Interestingly enough were the highest scoring design in Exclusion 1, by JR, the only of the high scores that did not make a MS2 gate.

Score 87%, cluster count 85, fold change 8.67

Vinnie’s design that are second highest scoring, there are two turnoff sequences and they work in a separate fashion.

Score 85%, cluster count 41, fold change 7.74

There is simply nothing anchoring to use for the MS2 turnoff that normally also often is used as aptamer turned off, to catch the FMN furthest away in sequence.

The MS2 turnoff and the FMN sequences are not really linked together in space. The MS2 turnoff is only connected to the closest by FMN. There seems to need to be a line of connection between the elements that one wants to bring together.

Exclusion 1 - Round 4  

As I am a fan of knotting the tails of the RNA - yes, nature do make small switches that switches in the open end (Example below)  - I made a series that attempted to make static stem out of the two ends of the RNA, despite knowing they would probably be too short.

I did managed to score 78% with one design and 9th highest score in that lab, counting only designs with 20+ cluster counts. But only Vienna says that the tail section folded, neither Vienna2 and Nupack said it folded, which makes it likely that it probably didn’t.

So as I suspected, making tails only pays if they are long enough to actually stay together.

Also despite I think bigger designs benefit from tucking away the switching inside themselves, it doesn’t help if the tails are too short to add stability. This lab wants dangling ends. :)

The highscorers in Exclusion 1

Interestingly enough were the highest scoring design in Exclusion 1, by JR, the only of the highscores that did not make a MS2 gate.

Score 87%, cluster count 85, fold change 8.67

Vinnie’s design that are second highest scoring, there are two turnoff sequences and they work in a separate fashion.

Score 85%, cluster count 41, fold change 7.74

There is simply nothing anchoring to use for the MS2 turnoff that normally also often is used as aptamer turned off, to catch the FMN furthest away in sequence.

The MS2 turnoff and the FMN sequences are not really linked together in space. The MS2 turnoff is only connected to the closest by FMN. There seems to need to be a line of connection between the elements that one wants to bring together.

Internal loops and switching

I also found it really interesting that the design has a loop in the “static end of the aptamer. As Omei have earlier mentioned that he noticed that loops were accepted 3 base pairs away from the aptamer at the static end of it.

Even more interestingly, Nando does the same as his robot and have a bit of the switching dangle placed in the loop. :)  Perhaps those strong internal loop dangles enhances the switch mobility. Jieux and Nando continues that trend.

The internal loops, reminds me of a discussion with Omei. We have been talking about the top end of the aptamer and its closing sequences.

Omei: There have been several cases where adding an interior loop to the neck, three bases distant from the aptamer, has increased the score. The base pairing of the closing three bases was identical, but the backbone wouldn't have been fixed as rigidly.

Here is one of them:

Xeonanis - 20 Mod55, Score 99%

Omei: If the closing stem is locked tight in a way that doesn't easily allow for the loop bases to take their ideal places, the FMN binding isn't going to be as strong. So in any place where the shape of an interior loop is important to the switching, there's a danger of making the backbone of an adjoining stem too rigid.

Strong stems (strength coming from both the number and the type of pair bonds in the stem) make the entire backbone (including the end base pairs) more rigid. If the backbone is too rigid at the stem end, it is limited in how much it can adjust to the specific ideal shaping of the loop.

For example, the FMN aptamer loop is most stable under some (unknown) location of each base. But to get into these positions, the backbone has to turn in specific ways.

Eli: Which reminded me that I had seen designs with the exact same placement of the last 3 base pairs at top of the aptamer and in some cases those even seems to split open in one state. I have a small section on it here.

Also I was reminded of some designs that were partial moving designs but still had full moving aptamer. I wrote a small section on them here.

Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
I think I have started coming up with different words for neck, ("strong fork" included also) to get it to reflect the long range crossing over of strands and multiple of them being present in the same RNA puzzle.

Neck (orange box) - when far away strands pair up. Usually with a multiloop between it and the further branching. The ribosome has several levels of necks on top on each other.

Yellow boxed stems are different from usual necks. I think the neck is central for closing of a section of RNA. When first that neck stem has formed, this extra kind of stem more easily happens. The neck is setting the scene. Knotting the section off. Anything stem forming of strands close in space hereafter goes easier.

Photo of whbob


  • 204 Posts
  • 67 Reply Likes
When I first heard the term "neck" used for the 5' and 3' ends pairing, as in a "neck" of some length creating a "stem", it made sense to me.  Like the open neck of a bottle:) 
Photo of Omei Turnbull

Omei Turnbull, Player Developer

  • 1008 Posts
  • 324 Reply Likes
Good point, whbob.  As far as I can remember, it made immediate sense to me, also.
Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Bottle neck, I like that one too, whbob. And now I don't have to imagine turtles with too many heads, although those with two are cute too. :) A glass flask can easily have multiple necks.

Also it reflect the nature of the neck. It really is a sections bottle neck. If the neck forms, the rest of the stems in that area that the neck holds, are more likely to fall into their right place. A better fit between the term and what it covered was what I was after.
Photo of salish99


  • 295 Posts
  • 58 Reply Likes
It would probably depend on the structural strength of the neck, I suppose. If it opens up during switching, it could be an open neck (scissor neck). If it stays the same, a close neck (bottle neck).
Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Exclusion 2 - Abbreviated MS2 - turning off MS2

Really when thinking about it, the MS2 turnoff stretch before the MS2 is actually an abbreviated MS2 with mainly the strong magnet segments kept. It’s kind of a double magnet. So instead of just the usual CC bases targeting in the primarily pyrimidine MS2/aptamer turnoff sequence, which is targeting both MS2 and the FMN, there is added an extra bit before with a G magnet - specifically targeting the MS2 C’s also.

This double magnet sequence is already present in Parushev’s original

The trend is continued in all the 4 designs with a score at 94% or above and with +20 clusters in Exclusion 2, round 4.

Mat’s high scorer

So the better designs tends to have a larger portion of the MS2 captured by the turnoff sequence, (right image, grey section). This also is happening in some of the high scorers, but non winners, but they tend more towards a long and full pair up, whereas the winners breaks some of the connection with the MS2, by adding in a mispair, like here a 1-1 loop. (base 34-53 in above image). So it seems like even the MS2 can get too much love at once. I think it needs it split up in portions/sections if the design is not to stay in one state too much. Else it will simply get too stable to get moving.

There is an interesting difference between Exclusion 2 and Exclusion 3. Exclusion 2 we have had a harder time making winners for than Exclusion 3.

For now it seems to me, that in the Exclusion 2 lab, it is harder to turn off the MS2 than in Exclusion 3. Exclusion 3 lab winners don’t seem to need to have 3 stretches of the MS2 covered for a turnoff (although one design almost do have full MS2 coverage happening). There are several winners there doing well with one (and shorter) MS2 turnoff that is targeting both MS2 and the aptamer on shift. But it isn’t needing to hold mirror sequences for both MS2 magnet segments.

So there is something about the switching in Exclusion 2 that makes it harder than the switching in Exclusion 3.

Exclusion 3 winner 100% (round 4)

Notice only short MS2 turnoff, targeting the MS2 G’s

Exclusion 2 winner 97% (round 4)

I have earlier said that it seems that the MS2 turnoff sequence much rather be on the right side of the MS2 (later) than the left side (early). There really seems to be something to it. :)

Perhaps this is also the explanation of the big difference in fold change between the Exclusion 1 and Exclusion 4 lab. Exclusion 1 lab gets lousy fold changes compared to Exclusion 4, and it has the turnoff sequence before the MS2. Exclusion 4 gets far better fold change and scores, and it has the turnoff sequence after the MS2. However there is one thing more. I really think this could be it. This is something that have had me mystified, why they were so different.

I think the turnoff sequence simply prefer moving in one direction over another.

Bigger end loop - easier switch?

One more thing. In Parushev’s design and the later mods of it, the turnoff sequence isn’t right next to the MS2 sequence as usual. But actually keeps a few bases distance. I find that interesting on its own. Perhaps having a bigger end loop form between the MS2 and the turnoff sequence is better than the tetraloop that is sometimes forming. From what I have heard, a tetraloop is more stable than bigger end loops. And we don’t just want stable here, we want switch. :)

Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Exclusion 3 - round 4 analysis continued

I did the same experiment for Exclusion 3, as I did in Exclusion 2 with aiming for opposite FMN. In contrast to the Exclusion 2 lab, the Exclusion 3 lab already had a good portion winners in round 3. I took the highest scoring Exclusion 3 design from that round 3, by Salish who was modifying Brourd’s design. And then I shifted aim with the turnoff sequence and shot for the opposite FMN sequence, the one furthest away from the MS2. I got this rounds highest scoring design that way.

Score 100%, cluster count 29, fold change 27.27

Now last round had an equally high scoring design, but it was better on both cluster count and fold change. So I can't fully rule out making MS2 gates for good.

Score 100%, cluster count 67, fold change 28.97

Omei’s design from round 4 has a higher fold change than both Salish and my winner and is following the pattern with aiming for the FMN furthest away from the MS2 = no MS2 gate, but it also have slightly higher error rate than what we usually set. Something I missed when I mentioned it in my first batch of round 4 analysis.

Omei introduced the idea with the 1.5 error rate limit. (How to set error rate limits in fusion tables) I sometimes lower it to 1.4, if there are loads of winners.

So at least for now, it looks like in Exclusion 3 lab, that winners can be equally happy with aiming for both FMN’s. But I still bet there is one preferred side. Or that it at least will end with at least one of the Exclusion 2 or 3 labs will show a clear preference. There might be something going on in relation to which way the RNA likes to switch. I think switching one way or the other do matter. I think each are not equally easy.

I think that to make or not to make MS2 gates. That's the question :) In some labs and cases they are still appropriate. Rather I think this will be like a handle to be adjusted to the situation. Just other of the RNA rules we use to design by.

Background articles

Which FMN sequence to catch?

Parushev’s design

Description of easy way to switch FMN target aim

Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Brourd’s mod of Exclusion 4

Brourd made the only winner in his lab.

Score 94, cluster counts 61, fold change 19.9

It has the MS2 turnoff targets furthest away FMN. 4 of the top 5 designs aim for opposite FMN.

In first round of this lab, I had one of my designs have its MS2 turnoff aim for furthest away FMN and it was the one of my designs that scored highest. It came in third highest, looking at designs that have minimum 20 clusters.

Score 84%, cluster count 79, fold change 7.29

Our Eternabot does something even more interesting. It targets both FMN’s, with a prolonged version of the MS2 turnoff sequence, instead of just one. At the same time it is getting really close to making a switch winner. :) Ok, I think aiming for both FMN’s at once is a strategy that will be normally best preserved for really hard labs, like the Exclusion 1 and Exclusion 6.

92%, cluster count 31, fold change 13.37

Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Exclusion 5 & 6 commentary

I originally proposed the Exclusion 5 lab, because I saw that MS2 benefited from a middle position in the sequence and from being between the FMN sequences. So far so good.

What I had not understood fully back then, was that labs that MS2 turned on in 1 state (turnoff labs/exclusion) were fundamentally different from labs that have MS2 turned off in 1 state. (turn on labs/Same state).

So what I proposed for this lab could only have worked for a turn on lab, but not an turnoff lab. The result was that it was hard to make winners.

However it and its sibling Exclusion 6 did taught me something valuable. They showed that turnoff labs are in particular need to have the switch elements in close range for them to be able to switch.

Exclusion 5 puts its MS2 further from the aptamer, than the earlier Exclusion labs. Exclusion 6 is the extreme case of Exclusion 5. But for some odd reason it is easier solving the Exclusion 6 that the exclusion 5. Perhaps the extra spacing also give some extra spacing to go full moving switch, something that a closer switch can not afford.  

Exclusion 5 - round 4

I have found that when something shows a skewed pattern to what is usual for labs that solves easily, then it is worth noting. Something that do hint that indeed Exclusion 5 is different from Exclusion 6 and the other labs too, is that the high scorers (in the early 80’es) do employ a raised amount of U bases compared to normal.

Here is the round top scorer by jandersonlee’s.

85%, cluster count 53, fold change 7.55

Notice how pyrimidine is across pyrimidine in the loops and similar purine is across purine in the loops. Whereas they pair in the stems. Makes sense. But how little the bases are mixed are more intense than usual for a static RNA or for a partial moving switch.

I decided to check out the U percentage out in the fun columns that Meechl added in her spreadsheet. And really Exclusion 5 has a raised number of U’s.

Most of the Exclusion 5 don’t score high, so I have set score limit to 70 to ensure to get at least some of them along.

Actually this pyrimidine repeat pattern reminds me of the early switch days and designs showing repeat patterns. Long stretches of pyrimidines to bind up the aptamer sequences and purine repeats as well. The labs that did bad then, had many switching base pairs in their structure by design.

Periodic repeats

Different types of switches

Partial moving switches also shows repeat bases - in particularly in the switching area - however it seems that full moving switches shows even more.  

For Exclusion 5 I had placed the switching elements, MS2 and the FMN, so they got too far from each other. (4 bases in this case) That meant that the stems or base stretches placed between MS2 and FMN could easily get too long and stable, in the way and thus counter a switch.

What my Exclusion 5 laboratory started to do and what what Exclusion 6 completed, was to force solvers to make their switches moving and move big time. I on purpose tried solve in partial switching style, which brought in a line of 0 scores and 30. (Link)

I think the only way of solving these labs - well in particular Exclusion 6 is to go full moving switch.

Another thing I noted was that many of the high scorers keep the general structural pattern. Which in this case is two halves folded together, with some loops in between.

Exclusion 6 - Round 4

If the stretch between MS2 and aptamer gets too big, so one MS2/aptamer turnoff can not reach to do the job on its own, an extra turnoff sequence tends to get added. I have earlier seen full moving switches use double turnoff stretches.

Jandersonlee’s winner made over a ViennaUTC design, did exactly that.

96%, cluster count 30, fold change 19

Again the MS2 turnoff is next to the MS2. This is generally seems to be the case for exclusion labs. The need of the MS2 trumps the need for turning off aptamers, which kind of makes sense as MS2 are the much stronger party.

However, if the MS2 turnoff is on the left of the MS2 which seems to be the less efficient side for a turnoff, then it may make a few bases distance to the MS2.

Notice that the pure aptamer turnoff (targeting FMN1, is far shorter than the MS2 turnoff sequence. The aptamer doesn’t need as long as a turnoff stretch as the MS2, as it is weaker and easier to get open than the MS2.

This design is kind of more open and split in half. Is this normal for switches that need more persuasion to move? Yup, I think it is.

Most of the high scoring designs are mods of ViennaUTC’s Quisina design. I find it very fitting that the bots are doing quite well on these more messy close to full moving switches.

Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Length of aptamer gates - depends on hardness of lab

Aptamer gates rarely goes beyond 6 bases as already 5 bases makes a pretty stable stem that it takes something extra to get moving. Switches are not friends of longer stems being involved in their switching area - and meant to be breaking open and switching.

Even for the stems meant to be static in switch designs, start benefits from GU’s and small internal loops, if the stem region is longer than 7 base pairs.

Static designs are fond of stable stems. If one is not careful with the sequence used for a 3 base pair stem, it can easily go unstable. Similar for 4 base pair stem. But the longer the stem, the more effort it takes to mess it up, so it does not fold. Which can be seen in some of our past static labs with really long stemmed designs. Many of the top scorers holds the most hideous patterns inside the stem, that would only fly and stay stable due to being hidden away inside the long stem.  

Score 87%, cluster count 85, fold change 8.67, Exclusion 4 top scorer by JR

Here the MS2 is unusually just trapped from the one end.

Third highest scoring design. Notice that not only does it has a hideous long aptamer gate (7 bases) but it even have two GU’s and crossed. Those act as splitters. So the long aptamer gate is countered by something to get it moving. And I think the longer aptamer gates for these rare Exclusion 1 highscorers are tell tale sign that this lab is hard and can’t be solved as easily by normal measures. What would work in Exclusion 2 and 3 doesn’t count here.

81%, 31 cluster counts, fold change 5.35, Nando

The second high scoring design takes a different route which employs two other tricks for getting things moving when a design are hard.

First it adds a loop 3 base pairs away from the “static” end of the aptamer and have it open in the switch, something that Omei noticed:

Internal loops 3 basepairs away from the aptamer end

This gives a bit more moving room for getting the switch happen. Rationale: When the switch don’t have strong switching incentives in the one end of the aptamer holding the MS2 sequence - where they really needs to be - add some in the other end as well.

Second, it doubles its amount of turnoff sequences from one to two and use one to target MS2 and FMN1 and the other to solely target the FMN2 sequence. This technique I have seen in play in other of the harder switch labs, like Exclusion 5 and 6 and more in 6 that is hardest.

Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Sum up of what I have learned from this round of labs

  • Mostly MS2 needs to be held from both ends - in the state where it is not turned on.

  • Multiloops in the switching area can get too big.
  • If the RNA end sequences are uneven in length, pair them up and make a tail or a hairpin of the extra bases - to remove excess bases from the main design - that would otherwise easily prohibit MS2 and FMN being close enough in space and sequence for a good switch.

Hardness of labs:

  • I think pressured MS2/FMN switches are labs that have their switching parts placed either too close or too far apart.

  • Lab designs with the MS2 turnoff sequences at the right hand side of the MS2 are easier to solve. (Exclusion 3 and Exclusion 4) Turnoff sequence to the left of the MS2 makes the lab harder to solve, compared to the similar but reversed case. (Examples Exclusion 1 and Exclusion 2)

  • Designs that have the MS2 turnoff sequence before the MS2, benefit from having it targeting the part of the FMN sequence that means it avoids making a MS2 gate. Designs that have the MS2 turnoff after the MS2, can aim for both parts of the FMN sequence and don’t get so much trouble of making an MS2 gate. Generally FMN1 seems to be the best FMN bit to aim for.

  • The length of aptamer gates are connected with the hardness of the lab. Harder labs tends to get longer aptamer gates. What makes a lab hard depends on which side the MS2 turnoff is put at in relation to the MS2. The length of aptamer gates are also related to if the MS2 is bound up to the rest of the design, from both sides.

  • Switch labs can have too little pressure on them, like not being connected well enough to help the switch elements move in relation to each other. (Example: The open ended switches Exclusion 1 and 4). I think MS2 likes to be connected to the rest of the design - and from both ends - preferable by backbone (Exclusion 2, Exclusion 3, Same State 2) and second best by hydrogen bonds. (NG 1 and NG 3, both lab types).
Photo of salish99


  • 295 Posts
  • 58 Reply Likes
Ah, so there is a new page. Well, here we go again, for reference, I posted the R97 analysis in the wrong getsat section:

Photo of salish99


  • 295 Posts
  • 58 Reply Likes
To emphasize what I am trying to do, I went a little nuts on the data crunching. After having filled up every single column in excel, I haven't really found some perfect way to emphasize a predictable trend.
To show you how some of the crazier ideas turned out, see:
Where Cx is the concentration of base A,C,G,or U.

I  got rid of the tent like structures and not have a defintive upwards trend in the 0.1 through 0.29 region and a sharp downwards trend in the 0.4 and larger region
Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Amount of switching stems per lab

How many switching stems a lab design needs are dependent on the hardness of the lab.

Regular easy switches

Generally 1-3 switching stems regularly fly. 2 is generally best for a normal easy to solve lab (Exclusion 3, Exclusion 2 and Brourd Mod of Exclusion 4 - listed by hardness from easier to harder easy). Ok, I have kind of been counting the MS2 as one stem all in all, but it is really two stems. JR already mentioned a strategy for stem length. (Link)

What is needed is the aptamer gate to form (State 2) The MS2 hairpin to form (State 1), the MS2 to get turned off (State 2) and the aptamer to get turned off (State 1) Which leaves 2 switching stems. It can be done with less in one state and with more. So the state holding the MS2 needs allowance for one extra stem.

Score 100%

Open ended switches

The more open and harder the switch - the better chance of fewer switching stems. JR did a top scorer in Exclusion 1, that only had one switching stem in one state.

Half open designs generally should be half stable switches. Notice how the part after the aptamer is static. If the lab is real hard like this one, there tend to be some allowance for having a small bit after the aptamer switching. Something Omei noticed.

Switching stems highlighted

Score 87%, cluster count 85, fold change 8.67, Exclusion 1 topscorer by JR

Pressured switches

The more pressured the switch - the more switching stems. Too much and wrong distance between the switching parts are from each other - in a way that makes it hard to bring them together - the harder to solve and the more switching stems are needed. Worst case scenario - all of them. :)

What makes a Exclusion lab pressured is different from what makes a Same State lab pressured. Exclusion labs reacts to distance between MS2 and FMN, Same state reacts both to being too close, and probably also to being too far away. I think the latter is why we are having trouble in our Same State NG2 labs - when compared to the huge amount of winners we managed to make in our earlier Same State 2 lab. We need to fold more of those excess loop bases into static stem to get better scores, to bring the MS2 and FMN closer in space.

Here is jandersonlee’s winner in Exclusion 6. It is the only winner in the lab. Notice that all the stems are involved in switching. No strand has the same partner between states.

96%, cluster count 30, fold change 18.66

Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
EternaBot Report

First a congrats to our EternaBot for almost making a switch winner! Note, I am still only counting designs in, that have a +20 cluster counts. :)

Eternabot took a different approach in Exclusion 6. It made two turnoffs separately - one for the aptamer and one for the MS2. Also it uses the much rarer G magnet turnoff for the MS2. In the first XOR puzzles the GU and CU turnoffs were both frequent, but as soon as we got lab data back for MS2 switches, the bias turned fully to the CU turnoffs. I think their advantage were that they could target the aptamer also, unlike the GU turnoff's.

Score 92%, cluster count 26, fold change 14

Fittingly enough the bots in general tend to do better in the labs that demands full moving switches like Exclusion 6 or 5 are hard to solve. Also very fitting a human came out on top - more specifically jandersonlee.

What EternaBot does right in the harder labs (Exclusion 5 and 6) - switching too much - is what it does bad in the easier labs. (Exclusion 2 and 3).

I can see there are some things our bot needs to learn better in the exclusion labs. So I decided to write a bot report, were I have collected a number of points where I think EternaBot can improve its behavior.


EternaBot is not very good at knotting it’s tails. So it has problems when the tail ends should meet. Although it is doing far better than its first round run on switches. Proof that most of the best scoring exclusion labs wishes for their tails paired. :) And they stay paired in both states. (These winners are mainly from the easier labs, except Exclusion 6 except)

Based on Meechl’s spreadsheet with additional fun columns added in.

Only exception among the high scorers is jandersonlee’s winner in the Exclusion 6 lab that goes full moving switch. Something that I in the case of Exclusion 6 think is necessary to even get a chance of getting a winner. In labs that needs go towards full moving like Exclusion 5 and 6, having the tails paired up, starts to get counter productive.

These labs needs to switch as much as possible to even have a  chance to get their switching elements close in space for them to interact. So knotting the tails there, will make for a bad solving strategy. (What not to do to solve designs that needs to go full moving)

Contrary to player design strategy for winning designs, EternaBot in general keeps too much loose dangling tail with too strong bases hanging around but not pairing and with no aim. Well except in some of its high scorers where it does knot them. The tail ends of RNA’s should be knotted as soon as there are the slightest possibility that they are long enough.

The aptamer generally needs a static stem at the end that is not having the MS2 sequence. Although Omei have noticed that some designs keep a loop 3 base pairs away from the aptamer:

Internal loops 3 basepairs away from the aptamer end

However EternaBot have some trouble with making that stem at the supposedly static end of the aptamer, static. A hairpin loop should also be made when the aptamer needs a static stem that has a long stretch of bases between it, where at least some of those bases needs to be a stem. Sometimes these end stems bend too much or they are actually switching with themselves.

Exclusion 1 and 4 are too short tailed for knotting. But the rest needs knotting. Well except Exclusion 6 and 5 that tends more toward needing to be full moving switches. Which leads to skills needed to be learned.

I have already stated made a strategy for when to knot tails. But I think I can add some specifications.

Sum up on tails

In easy labs - always knot the tails. In hard labs, avoid to knot the tails.

Further hint for improvement: Knotted tail sections should be equally long or close to equally long in both states - EternaBot tends to make way too many bulges and have too many single base hanging loose at its tails.

To MS2 gate or not - Exclusion Labs

I have earlier submitted a strategy dealing with MS2 gates. But I can see that my strategy for FMN targets may have gotten the bot confused. EternaBot too often targets both FMN’s, as my strategy rewards both.

But I think I have a better idea now what is working better.

Also while I was cautious of ruling out the MS2 just yet, after watching the data across several labs, where several of us have tried to target opposite to usual FMN region, I see do see that MS2 gates in general are starting to disappear. They won’t disappear fully. We will still need them. However I’m starting to see a pattern of in which labs they of less help compared to which labs where they are tolerated.

Some labs are are fifty fifty about what they want. Exclusion 3 is one of them. The MS2 turnoff sequence can target either FMN sequence also and we can still still make winners.

In Exclusion 2, it matters which FMN to target, it is easier to solve it if aiming for the FMN that means no MS2 gate gets created.

So what mainly decides if there needs to be a MS2 gate or not, is the position of the MS2 turnoff sequence in relation to MS2. If it is before the MS2, the lab is per definition harder to solve and it will benefit from avoiding making a MS2 gate. If the MS2 turnoff sequence is after the MS2, then the lab cares a lot less what method is used. MS2 gate or not.

Static aptamer end

Also EternaBot is not very good at keeping the end of the aptamer static either.

In easy labs, the end of the aptamer that is not holding the MS2, should always be the one knotted to become static. (Getting the switching parts together) It may be less static and even contain a small internal loop some 3 base pairs away from the aptamer as Omei noted. I think this should mainly be allowed in the open ended labs. (Link) EternaBot is often not able keeping its static aptamer end, not switching.

Look out for optimal aptamer end closings for the first 3 base pairs. Not every combo is equally effective. Copy the average from winning designs.

EternaBot also have a tendency of bending the stretch after the aptamer end that should be static too much. It’s good to put in an internal loop and or a GU when the stretch after the static end of the aptamer gets longer than 6 bases, but too much bending of the stems at the static end of the aptamer isn’t helping. Only in the open labs it is an advantage to make stretches at both sides of the aptamer get involved in the switching.

So there are two kinds of full moving behavior that can be in switches. Full moving switcheswhere all the stem parts move between states, and then there are full moving aptamers where both regions around the aptamer is involved in the switching.

The latter can be separate from the first, while the first can't be separate from the latter.

So full moving aptamer can turn up in harder to solve lab as a way of getting things moving a bid to aid the switching. In easy labs, such trickery generally isn't necessary. If the lab gets totally hard - because the switching parts are too far apart in sequence and can't get brought close in space - then it needs to go full moving to even have a shot at solving.

Aptamer gates are too long

EternaBot is making its aptamer gates too long at the end of the aptamer that needs to get moving in the easy labs (Exclusion 3 and 3) Whereas the aptamer gate do need to be longer in the open ended labs.

Multiloop too big - make static stem

EternaBot’s multiloops are too big. Too big multiloops tends to be a switch “turnoff”. :) There shouldn’t be much more than 3-6 single bases in the multiloop ring in between the aptamer gate (the one not next to the MS2) and the static stem and again between the static stem and the MS2 turnoff sequence. Sometimes even less will do. (Some of these ring bases will be a 3-4 base Salish hinge - consisting mainly of A bases, but sometimes mutated to other bases like U or G - this is one of the heavily mutated regions.

A way of dealing a way with the excess bases are to make a static stem. Which brings me to that EternaBot is not very good at making its static stem in the switching area. The static stem should be somewhere in between the aptamer gate and the MS2 turnoff sequence. Which it should be closest to, tend to depend on which side the MS2 turnoff sequence is to compared to the MS2.

There is typically a Salish hinge either between the aptamer gate and the static stem or between the static stem or the MS2 turnoff. Which it is also tends to depend on the position of the MS2 turnoff in relation to the MS2 sequence.

Color of the MS2/aptamer turnoff sequence

I think EternaBot regularly tends to get the MS2/aptamer turnoff sequence wrong. Which is probably due to my earlier strategy on it.

The MS2 turnoff sequence is often too reddish - whereas we with time have learned that a pyrimidine stretch is generally the most effective turnoff, while G’s only occasionally works also.

Distribute the GU’s

EternaBot tends to use a little too much GU.

Designs like Exclusion 5 and 6 that have bigger distance between the aptamer and the MS2, will need more GU’s to get moving. So harder switches takes more GU.

There is something like different GU percentages depending on states. The one state can usually take a little more than the other. Which it will be will depend on the kind of puzzle, how hard it is. The GU’s gets added on the state that is hardest to get moving. Like by structure means - like Exclusion 5 and 6, where the switching parts are too far apart in sequence.

Pressured switch labs

Any exclusion lab that have their Aptamer and MS2 sequence more than 1 base apart needs to start think about going towards fuller moving switch. Those types of labs can’t be solved with the normal strategies for making a partial moving switch. They are generally harder.

If a pressured switch lab design (One with its switching parts forced too far apart in sequence to get them close in space) gets solved with a partial moving switch strategy, it will not solve well.

Similar if an easy switch lab design gets solved with a strategy for making all its part moving, it is extremely hard to hit on the exact right sequence to make the switch full moving to get it solved. It is always easier making a few base pairs switching correct, compared to a huge bunch of base pairs. It is far easier isolating the switching area in a small confined area and have stable areas on both sides. It is easier controlling a switch that way.

Same state labs gets stressed if the MM2 isn’t at least 2-3 bases away from one of the MS2 sequences. But turn on/Same State labs aren’t as easy to stress as Exclusion labs. So the strategy is mainly for the Exclusion labs.

Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes

I was at writing report on EternaBot to see if I could spot some things that could improve its score, and while I was at it, I decided to take a look at the player designs too. Basically with the same aim. :) We have done good, and I think we can do even better. So here are some things I think are worth looking out for.

I have summed up, what I think was a problem for each lab type. I hope this will be a help for understanding what can be a problem in a particular type of lab design and why.

A switch design isn’t just a switch design. Switches have personalities. They want specific things, just dependent on their structure. Also they don’t come as equally hard to solve. The easy labs takes a certain solving style that won’t work on the hard labs, and similar the hard labs takes a different solving style that won’t work on the easy labs.

I think if we use this knowledge, we already have very strong cards on the hand to tell what a particular FMN/MS2 switch needs, just from what its structure is. I strongly suspect that this will be applicable to other types of switches too.

Exclusion 2, 3 & Brourd’s mod of Exclusion 4 - Easier labs

Here I list what I think gives bad scores for this kind of easy labs.

I have included Brourd’s mod of Exclusion 4 in this series, as it does what the easier labs, Exclusion 2 and 3 does. Allow the switch to get locked away inside itself. I will just mention on which cases it sticks apart from the other 2.

  • End tails:

    • End tails not tied with each other = unstable aptamer end

    • End tails dangling but not involved in anything

  • Aptamer not locked from one end

  • Moving: Too much of the design switching - not partial enough switch

  • MS2:    

    • Attempts to bind up way too much of the MS2

      • Way too long aptamer gates

      • Way too long MS2 turnoff sequence

    • No MS2 turnoff sequence next to the MS2 sequence

    • Turnoff sequence not pyrimidine enough. :)

    • Turnoff sequence length too short or too long. Needed: 4-5 bases with a little more to Exclusion 2, because this lab is harder due to the left positioning of the MS2 turnoff. But turnoffs that get more than 6 bases needs splitting - else I suspect they make the MS2 stay turned off and not capable of switching fast.

  • Structure:

    • No static stem made of excess bases

Not needed for Brourd’s mod of Exclusion 4, as there were no excess bases. Which is actually kind of interesting given that we have been easier able to make higher scores in the other two labs that allowed space for excess bases. However this Exclusion 4 lab still do have the disadvantage of having the MS2 and FMN 1 base apart, which Exclusion 2 and 3 don’t. So I can’t say which is which.

    • No Multiloops, many of the lower scoring designs have no multiloop. It is the natural consequence if the tails are not knotted with each other.

    • No salish hinge (3-4 base sections of mainly A bases spiced with an occasional U or C, located in the multiloop in the switching area. Should be after the aptamer gate strand that does not have the the MS2 next to the aptamer.

Exclusion 2 needs the Salish hinge less than Exclusion 3, namely because Exclusion 2 needs a stronger MS2 turnoff. But I’m guessing it will benefit the results, should we have a few extra bases space for getting it in.

    • Turnoff sequence not pyrimidine enough.

  • Stems:

    • Too long stems, 6+ and above is too long.

    • Too many switching stems. Exclusion 2 lab is harder than Exclusion 3. This lab benefits from 2-3 switching stems, normally I will only recommend 2 (Counting MS2 as 1). The reason for the extra needed stem, is that the MS2 needs some more help with getting turned off. Making the MS2 turnoff too long, is not working as well. I suspect it prevents the design from jumping states as quickly. It get too stable in one state over another.

  • Base pairs:

    • Too much GU’s. Fairly easy riboswitch labs mostly takes a % of 0-5 and sometimes up to 10% above that and it is not helping. Especially not if in the switching area. You may get easier away with it in static stems.

For some reasons Brourd’s mod of Exclusion 4 takes a little more GU than the more average easy switch lab. Trends are that the harder the lab gets, the more GU. I consider this lab harder than both Exclusion 3 and 2. One of the reasons for that is that it have introduced 1 base distance between the MS2 and FMN, just as the Exclusion 1 and 4 lab has. But I think this small distance already starts makes the lab goes harder, despite it has long enough tails to knot. Fairly easy riboswitch labs takes a GU of 0-5% and sometimes up to 10%. This lab can go to 10%.

    • Too much GC’s or too little. Generally switching stems should have one 1 GC pair as a minimum, but they don’t like too much more either. Often 2 is the maximum, unless there are special circumstances that does that the lab needs more. Basically how many GC’s you need per stem, depends on what you want a switching stem to do. If you want to turn off the MS2, you will tend to need in the high range of GC’s. To counter the MS2 hairpin forming, as it really want to fold. At least in Exclusion labs. So for the turnoff sequence you can regularly away with more than 2 GC pairs. Still it depends on the general hardness of the lab. Exclusion 2 is harder than Exclusion 3, because it has its MS2 turnoff sequence to the left, where it has weaker effect on the MS2, than if it was to the right. So to solve Exclusion 2 more GC pairs (forming between the MS2 turnoff and the MS2 sequence) are needed for the MS2 turnoff.

Exclusion 1 & 4 - Harder labs

Here I list what I think gives bad scores for this particular kind of lab. and what elements are missing.

  • MS2 not caught from both ends. JR had a top scoring exception in Exclusion 1.

  • Aptamer not locked from one end. Although open ended designs can contain a loop around 3 bases away from the “static” end of the aptamer, as Omei noticed.

  • End tails:

    • Attempting to tie up the tails as if it is an Exclusion 2 and 3 lab, is a bad idea - the tails are too short to hold the pressure from the rest of the design attempting to switch. But I had to try. :)    

    • Too long end tails dangling but not involved in anything

  • Moving:

    • Too much of the design switching - not partial enough.

    • Too static - trying to solve the design with too few parts switching. Absolutely not working. But I just had to try. :)

  • Structure:

    • Multiloops - happens when the tails are paired and these are not helping. As I suspect it is not holding.

    • No MS2 turnoff sequence right before the MS2 sequence

    • Turnoff sequence not pyrimidine enough. (Exclusion 4, not the case in Exclusion 1)

    • Aptamer gates - too short. Typical length 3-7 bases - longer is probably due to the lab being hard to get to switch its MS2 as it is not bound up from both ends.

  • Stems:

    • Too many switching stems. Open ended switches tends to needs fewer switching stems compared to easy to solve labs and pressured labs.

  • Base pairs:

    • Too much GU and too little. Open ended labs tend to allow a bit more GU’s. A healthy range is typically around 5-10%

    • Too much GC’s or too little. Generally switching stems should have one 1 GC pair, but they don’t like too much more. MS2 is the exception. So like this, you should usually have one GC in a switching stem, if it is short you may get away with 2 GC pairs. Though when countering the MS2 hairpin which really want to fold with it self you can get away with more than 2 GC pairs for the turnoff sequence.

Exclusion 5 & 6 - Hardest labs

I list what I think gives bad scores for this particular kind of lab.

  • End tails:

    • End tails tied with each other = too stable design and aptamer end. But I had to try. :) Result = Bad

    • Lack of tail dangles - but hard full moving labs seems to like them - contrary to partial moving switches

  • Aptamer:

    • Aptamer locked from one end. (Exclusion 6 does allow that in some of the higher scoring designs though). These labs tends to need to go towards full moving. Having too many static stems are counterproductive to be able to get a solve. Not that it did not stop me from trying it out. :) Result = Bad.

  • Moving:

    • Partial solving style is generally not working - these switches needs to move. :) But I still had to try. :) Result = bad.

(Exclusion 5 needs to move more than Exclusion 6)

  • Structure:

    • Multiloops are not helpful for making full moving switches move. They add way too much stability. :) Of cause I had to try. Result = Bad.

    • Static stem - should be banned. They prevent the switch from reaching its full moving potential. I did try. Result = Bad.

    • Not enough turnoff sequences, more are beneficial as one typically can’t reach both switching elements due to switch element distance. Often both FMN sequences needs to be caught by turnoff sequences. I did try with a single one. Result = Bad.

There needs to be at least 2 and sometimes 3 turnoff sequences. One for the MS2 and one for each of the aptamer sequences. The switching elements in a full moving switch does not always want to share turnoff sequence.

Exclusion 6 also benefits from a slightly more purine MS2 turnoff, compared to the more usual pyrimidine turnoff sequences. Actually it appears that this exact lab doesn’t mind the longer stretches of G’s and U’s, compared to the easier labs. The GU electronegativity and their not wanting to bind properties in especially switch designs, may be an advantage here, like oil reducing friction. To get the full design to move for making the sequence go unstuck and switching. (Penalize GU heavy regions)

  • Base pairs:

    • Too much and too little GU’s. The harder the switch, the more help it needs to get moving. Typically takes around 5-10% GU’s and sometimes more. Hard labs like these tend to accept more GU’s than the easier labs.

    • Too little GC pairs. Exclusion 6 take a higher amount of GC than usual in the switching area. (which is the whole puzzle) Exclusion 5 on the other hand takes more AU. I’m beginning to think that these two designs with their switching too apart, are quite different, despite sharing a lot of tendencies.

Extra comments on Exclusion 5 & 6

Exclusion 5, tends toward being first half of the RNA pairing up with the last half and then a small slide. Example of jandersonlee’s high scoring mod of Malcolm.

Score 85%, cluster count 53, fold change 7.55

Exclusion 5 lab benefits from having 2-3 internal loops in either state.

Exclusion 6, too much of first half of the RNA pairing up with second half of the RNA. Needs to be split in two sections, that switch with each other.

Jandersonlee’s winning mod of ViennaUTC in Exclusion 6.

Score 96%, cluster count 30, fold change 18.66

Photo of Eli Fisker

Eli Fisker

  • 2289 Posts
  • 518 Reply Likes
Molecular movie

I think what the MS2/FMN switches want are something very specific. Generally the switches tends to want the same structure and the labs that scores lower, I think does so for lack of those specific structures or for preventing them from being made. There are exceptions but the overall trend seems toward that some specific structure is simply best.  

Another reason for the structure similarity, is that many of the high scoring designs are made over sibling designs. But I don’t think the one exclude the other. That the high scorers want something specific and thus being very alike.

When I sort the Exclusion labs after score and fast forward through them with Nando’s “D” short cut, then the RNA’s tend to have quite alike structures for the highest scoring designs and go far more kicking and moving, the lower the score goes.

I think where the shape of the structure could change the most, are for those labs that goes full moving, like Exclusion 6. And even the Exclusion 5 and 6 have quite much similarities for the high scorers in their individual labs.

Most specific in structural wants, is the FMN bound state

Another interesting thing. The 2 state seems to change less than the 1 state. This seems to go across the labs. There is generally more structural changes in the 1 state.

However it kind of makes sense. When the aptamer is folded in second state, it isn’t a part in a multiloop that can vary in size, as it is in State 1. Kind of makes sense that the state that hold the aptamer will be the least variable.