Let's get started!: http://www.eternagame.org/web/lab/689...
This is it! :http://www.eternagame.org/web/lab/689...
Initial, preliminary data are available for some puzzles in the first round.
Google Spreadsheet: https://docs.google.com/spreadsheets/...
Excel file: https://drive.google.com/file/d/0B_N0...
For now it were some fusion style designs that did best for the hard labs. Kudos to Atanas for showing that our sub lab designs really could be added up for the hard labs. :)
Other patterns have started to come through also. Intersection - where an input is split up and bound apart in sequence has started to grow more frequent with a higher amount of inputs. This intersection thing I first recall seeing in some Eized mods I described here.
Another thing. Many of the High scorers have part of their sequence hidden away when in a state where they don’t need it. This is something that can be useful to do in combination with intersecting inputs.
And then there is seriously inter-entangled designs.
HARD DEC LAB
Funny recurring MS2 like hairpin
Several of my High scorers (a fusion of a Malcolm and jandersonlee design) have something like a MS2 hairpin in the fold connected with the reporter. Something I have noticed in several designs both while designing and looking at others. This is something I have found amusing. The first round reporter sequence in particular even shows great likeness to the MS2 sequence. Although bending opposite.
Legit MS2 hairpin
The MS2 ghost hairpin is made of the reporter and part of the nearby C complement. Hereby inputs are used used directly for a turnoff against each other by a nearest neighbor strand pairing.
The reporter is being countered by a nearby “anti-reporter” sequence. A MS2 turnoff as I would have earlier called it when we used MS2 for reporters. In this case it was one of the input sequences that fitted to do the trick. But an unrelated but matching sequence would do the trick too.
HARD INC LAB
B complement moonlighting as both switch turnoff and turn on
Half the B complement is used for turning A and R off. The other half of the B complement is overlapping with C and aiding C get turn off when C concentration is good - in other words when A and B concentration is not high enough to counter the concentration of C.
The highest scoring design uses a sequence overlap between C and B for tipping between state 1-3 and 4. B and A are turned off by an overhanging C end tail that targets a G stretch in the B complement. So when C is strong and when A and B are weak, the BrA stretch is binding with itself and turning itself off.
Simplified structure drawing showing inputs against the different states.
Basically one half of the B input is working as either state 3 turnoff or state 4 turnon.
Intersecting style in action
As can be seen in state 3, the C complement that is being involved in the switch is actually split up. In between and hidden is R and its anti sequence (plus some of the A complement too).
I have earlier mentioned that I expected intersecting elements to grow more prevalent the more inputs we got. This seems to hold.
Global fold change versus score
There are no design with high fold change. Actually global fold change is generally higher than fold change - which is a new. :)
For these two hard labs all designs gets fold change below 2. So most of the designs don’t switch really well. But we have a fine starting point for improvement.
While several of the fusion design get in the 10 of global fold score, what do get real high global change is a more inter-entangled design style and short distanced interaction of inputs.
For now it seems that in most cases when we raise global fold change, local fold change suffers. And when we raise local fold change, global fold change suffers.
Things that seems to help switching get going
Shorter binding stretch for the input.
Some of the smaller labs with just one reporter and one input or two inputs, seemed to get grumpy if the whole length of each each switch element was used.
We are using a lot of the input sequence for many of our hard tb lab designs and we may gain some momentum if we shorten things down.
Things that seems to help raise global fold change
Direct interaction between inputs/reporter
Simple interaction: Lane sharing with direct overlap between the sequences of inputs/reporters (2-3 inputs)
Short distanced interaction: Jumping jack style pairing where inputs/reporter bind directly with each other for turnoff (2-3 inputs) Inputs can be spaced by a static stem.
Intersecting of complements. Splitting up input complements/reporter and spaced by another switching sequence.
Inter-entangled style. For a design with 3-4+ inputs - long distance 3D interaction between inputs/reporter or outside sequences starts to become beneficial.
Several of these can be inter combined.
Key point on global fold
Here is what I have seen help global fold change up til now.
The more directly you get the input complements to interact with each other - the higher global fold change you will get.
There are different ways to do this and what to choose will also depend on the amount of inputs and nature of the inputs (eg if they fit well for a direct overlap (lane sharing).
1)Lane sharing with direct sequence overlap
Simple interaction - direct overlap will do - can be done when inputs are short and fits well for a direct overlap with each other. This approach will work well when there are few inputs and when these inputs fits well for overlap. (like 2-3)
This has yielded the yet highest global fold change in the R3 lab. (37%)
2) Middle man sequences
Using neighbor sequences to interact as middle man between the inputs/reporter. (Usually gets higher fold change but mostly is less lucky with global fold change.)
3) Sequence match in 3D - Direct pairing between inputs/reporter
This approach rely on making input complements/reporter pair up in 3D.
A separated but directly interacting inputs/reporter over a short distance. A distance could be created by a static stem or a sequence temporarily packed up in a stem. This former approach was successful in R2 for global fold (10%) and score (95%).
Here is the design with the absolute highest global fold change of the hard TB labs. I have marked the beginning and end of input A and B with black rings.
Global fold change 27%, score 32%
What I find interesting about this design, is that the A and B complement are directly connected with each other. But also even with the C complements. They have an option for direct feedback between the different parts.
The B and A complement are capable of turning each other off when neither of them are present and when C is. Thus Atanas got A and B multiplied. If A and B are present they will stop binding with each other. Now this design doesn’t score well. But I think it shows that when inputs are made dependent on each other they may start show the behavior that we are interested in.
Another approach that also gets high global fold change is making switch elements interact with switch elements or neighbor elements for turnoff in 3D. This was already seen in the logic gates designs. But has been taken to a more extreme degree. A longer range approach can be seen in Brourd’s designs.
Global fold change 24%, score 30%
Basically inter-entanglement is a sequence match in 3D.
Perspective for the future
So it seems that what is immediate of benefit to a single concentration - fusion style designs, wins out in the short run and for now. But I think the more inputs we get, a more inter-entangled style will win on a global fold scale. If we can just get it to unstick when needed.
I think we need think about we can play these different approaches for getting good fold change and global fold change together so we get the best from each.
For now I speculate if we can make these 3D inter-entanglement more MS2/reporter/switch element like, we should have more success making a switch. As I think certain sequences are having an important role in switching.
Imagine if C is the first attractor, but B comes along half way down the C oligo and continues after C.
This is like having pages placed on a desktop, each new page overlapping the previous page will dominate. The kernel attractor sequence can't be too strong of too weak to properly switch. It has to be "just right".
My problem is how to I determine what is "just right" :)
In the same puzzle, Mat747's designID 7094435 scored just below yours. Looking at it, it didn't pass the folding engine constraints.
It's things like this that make me wonder if I will ever understand what's happening, but I always look forward to your explanations to help me along :)
Winning strategy for now: Fusion style
I mentioned in the above post that I really liked that Atanas made the B and A interact directly with each other in the design that had the highest Global Fold this round. (32%)
However I failed to point out is that the DEC Fusion style 1.41 high scorer (73%) has a similar feature. While the A and B are not distanced by a static stem as in the case in Atanas design, they do interact in state 3.
State 3 and 4, B and A element borders marked by black rings
Similarly there are actually 2 potential binding sites for the reporter. (In state 3 in between the two B complements. Something which is more clear in one of Mats mods that did well. (68.8%)
Here is my earlier drawing of Fusion style 1.41, now with extra reporter binding site and A*B drawn in:
Here is the design with the extra reporter site highlighted in the design.
I decided to make a fusion with the elements in the same order as this rounds high scorer. So here comes a demo of fusion style aka addition Atanas style.
Here comes the parts that works for fusion (There are more possible combos)
First sub lab puzzle part
cB, cBr (cBA)
One of my sub lab submissions that had same order of elements and had the B complement split.
Second sub lab puzzle part part
I couldn’t find any of my designs or other denoted in this order, so I designed it denovo.
To fuse the sub lab designs, i took the first B/C part that should go first in the order.
I froze the design to be able to easily delete all the static stem bases in one go. Then I did a copy.
Then I pasted the sequence into the DEC puzzle. I moved the puzzle part fully to the end of the design (removed that first loose A base).
I froze the A/C design to delete the static stem bases in one go.
Then I slided the whole design to the beginning of the sequence to be able to paste it like I wanted.
After I unfroze it again. Sequence now slided to the beginning. Note that the structure is still stable while it does not fulfill all demands.
Then I did a copy, called for the sequence stamper and pasted the sequence.
I specified sequence landing space to 46
Total fusion after it has taken place.
Since the intersecting solve style fills more than a design that is just overlapping its element, I needed all the place for the fusion I could get. So I have slided both sub parts as close to the end as I could get.
Normally I would also slide the two parts back and forth in relation to each other, to find a potential match between them. But there was no space this time. (I would like to see a fusion of two intersecting style solved sub part, but mostly I think we would run out of sequence space for that.)
The main thing is that if we have a working structure from an earlier lab, it is regularly possible replicating and doing something very similar and have a shot of making it work. At least in simulation.
How matters the input to actual switching?
I have been drawing some of the majority types or solve styles present among the higher scoring A/C and B/C designs.
Now we have more data with the new reporter, I can say that the absence of MS2 and the presence of a weaker reporter instead, seems to allow more different solve types to happen. Still there are certain overall trends in the data. Solve styles that works better than others. I think that the MS2 in past labs had a habit of sparking certain patterns in the designs themselves.
What I find really worth noticing is that the winner in A/C DEC and the top scorer in B/C DEC do not solve in the same structural manner.
The A/C INC and B/C INC winners do have some of the same structural types, but do not have identical majority types.
Now why is that?
I have been a strong proponent of reusing the same structure, even for designs with different inputs.
I still am, however I have something to add.
While it will generally work well transferring a successful structural strategy from one lab to another and just refit the input - we have done so successfully many times - I also see another pattern start to occur.
I did structurally alike versions in all the Sensor RIRI labs - just with the input changing - however they don’t do equally well.
In both sensor A and C RIRI there are winners, however in sensor B RIRI lab the highest score was 81%
And I suspect this score difference is due just to their different input. As the input is the only thing that changes between them. Plus I nicked the starting structure from a winning design by Omei in a previous and similar lab - with an identical reporter!
Fold change and score
One thing worth noticing is that any lab that contains anything with a B input, trends towards scoring less well compared to the labs without it. (N.B. there are winners in several of the labs with the B input involved.)
Labs sorted by average score
Also fold change trends towards being lower for the labs having a B input, compared to those without.
Labs sorted by average fold change
The nature of the inputs
So why are the Sensor B RIRI’s and some of the other B containing labs doing worse than their peers?
I think the reason for this, is that the input will have some force over what structure is needed. This is what I'm starting to see and have suspected. Each input is not equal and alike. While many will be.
The B input had lots of strong bases and in an equal distribution. It was more stem like in nature, not being microRNA like. MicroRNAs seem to have a bias. (At least those I have seen) Like more G's to C's or reverse - but not a balanced mix. They also seem to not to be too happy about folding with themselves - practically a function of said bias.
So if an input gets too strong - having too many strong bases - compared to the other inputs, I think it will start affect things like fold change and score for the worse. Plus it should eventually be way too happy to fold with itself and also less willing to let go, when it first have attached to something other than it selves.
Perspective for switch inputs
Structure reuse between similar switch labs will generally work. But the input has some power too, dependent on its nature. If that nature is strong bases, things will get sticky. And other structural patterns will need to get played to get to a solve.
So basically I think the input alone is capable of sparking the need for a new kind of structural solve, if of a certain not so mellow nature.
It's fundamentally the same thing I think happened when it came to the change from the MS2 to the new reporters. But in this case the absence of the MS2, is releasing more options for solving structurally different.
Most of the labs I drew above use similar solving path ways, but they don’t get played in the same order, depending on what lab type and input is used. Yet a thing worth noticing is that the B/C DEC high scorer gets around the too strong B input by splitting it up in two and thus weakening it. A similar pattern can be seen in the ABC2DEC and A (AND) B - RI high scorer.
There are two designs in this lab series, one in Sensor A RIRI and in Sensor C RIRI each, with high fold change (200+).
What these two designs have in common, is that they both take advantage of a specific sequence stretch that is build into the input. In other words, something that is unique to that input for both its sequence and its position in the input.
In both designs this involves magnet segments. I think these aids the switch even if these switch also involves longer than usual switching stem area.
Thus I expect the structures for these solves will be harder to transport between labs, as the other sibling labs likely won’t have input that can satisfy these demands and may thus not allow for a similar solve.
However I find these designs very interesting. In that they may open the door for hitting high yielding switches for a particular purpose - by input sequence design. To get switches with high fold changes, we could could cut the input sequence in such a way that it gave us the wanted sequence element needed to make such a the switch more switchy. And even better if we already have a high yielding switch, as we can reuse the structure and get ideas from the input on which kind of sequence would be most successful.
Sensor A lab example, Score 100%, FC 210% (Nearest neighbor strand fold type)
Switching magnet stem formed in state 1, highlighted with green (CU rich) and red (GA rich). Which is exactly the kind of sequence I again and again see involved in switching elements.
All of the reporter is hidden away in a hairpin loop (State 1) between the switching magnet stem. Most of the input A is packed away too. Only a little sequence is “dangling” in an internal loop. (also state 1).
Sensor C lab example, Score 100%, FC 268%
Also here the reporter is hidden away in a hairpin loop.
Here is an example that also uses similar approach. It makes the R fold with itself. But also take advantage of that the input complement batches up with the reporter for its turnoff and the input for its turn on. (Colored boxes)
Score 100%, FC 192%
So this design structure above may be reusable in another lab, if the input has double C bases last or close to lase - meaning that the input complement would get double G as its first bases. (Another thing to take into account here is that the input seems to being able to slightly fold with itself - state 2 - 5’ tail. This may or may not be important for success also.)
Though I think the above design will be harder to make work in general with many different input types. Compared to Jumping jack style designs that seems to be more robust to changes. It is possible to make high scorers with different inputs and different order of input and reporter, just some lab designs suffer more in fold changes than others.
Basically using an input with similarity to an input that has already showed up in a winning design that has the lab conditions one wants, both replicating the design structure of the winner and the important parts of the input sequence (like magnet segments used or sequence area that has high switchability potential), will allow us to make more high fold change candidates.
It is not just the structure that needs to be the same across designs, to some degree, important regions of the input also needs to be the same to have a similar function. By designing the input by actively choosing it such that it resembles earlier successful inputs, I think it much will raise chance of succeeding again with a similar structure.
My main point is, that if a winning design in its input has a magnet segment (or an otherwise switchability enhancing sequence) at particular spot, which is involved in the switching area, then that design structure is more likely to be reusable in another lab if the input in this other lab also carries a similar magnet segment at a similar spot.
Structure overview of some common high scoring types. NB, there are more than usual. I think the this weaker reporter, compared to the MS2 reporter, allows for a lot more structures to be legal.
Jumping jack style designs
I mentioned that my Omei inspired jumping jack series and their derivatives got different scores, just depending on which lab they were in. So the same structure was general doing well in all labs compared to their mates, the designs did not do equally well on score or fold change. (not fully identical but overall similar structure). Something I think is due to their difference in input sequence.
Sensor A - RIRI, 93%, FC 35%
Sensor B - RIRI, Score 80%, FC 9%
But despite the B lab getting lower scores, the jumping jack style still got 3 highest score in that lab, something I think hints at that it is more a problem of the B input than the structure itself.
Sensor C - RIRI, Score 100%, FC 97%
Second best fold change in this lab.
Now these designs are inspired by Omei’s winner. They are not as the originals. Had they been, they would have likely done even better. I reversed the reporter and input in order.
Also there is one more thing that is not there at the same spot - is something that is in the sequence of the input in this earlier lab. The inbuilt magnet sequence.
mRNA-in, reporter-in, Score 100%, FC 354%
Now I did get the general design structure working (while reversing input and reporter), so it can clearly work without that magnet segment, I’m just sure it will work better with. :)
I still think jumping jacks style design will be generally easy to make work also with many different kind of inputs. I think they may be a little less input sensitive than other structures - although they will prefer a magnet segment too at a specific spot in the input sequence too. Despite they didn’t get that and that I reversed the order of the reporter and input compared to the originals, they were still doing fairly well in the RIRI labs.
So this is basically exactly the same story as the one on the high fold change designs just above. Just with a different design from an earlier lab. And even better fold changes.
May you keep your switching areas rich in GA stretches, CU stretches and if you need need to move a switch mountain of stems also a long region gliding GU stretches :)
I was making a prediction for the round 1 RIRO labs. I was expecting the Sensor C riro lab to behave better than the Sensor A and Sensor B lab.
I was basing this based on behavior I have seen in labs with designs with direct overlap between their input complement and reporter in exclusion labs and a hypothesis that it is better to have the leaving input before in sequence than the staying input.
The Sensor C did indeed do better (count = winning designs), with the most winning designs compared to sensor A and B. Both Sensor A and Sensor B managed to get winners too.
Now this overview don’t tell the whole story as some of the designs are of different kinds.
So instead I compare for the winning designs across these sensor RIRO labs that used partial overlap between reporter and input. Which are those designs I’m interested in comparing.
Sensor A - Two winning designs uses direct overlap in non preferred order (FC range 60-91%)
Sensor B - Five winning designs uses direct overlap in non preferred order (FC range 42-46%
Sensor C - Fourteen winning designs uses direct overlap in preferred order (FC range 39-205%)
What else happened?
Sensor A got the overall best fold change 281% for all 3 labs - with a non overlapping design - intersecting style. This design had several siblings, raising the overall fold change average of this lab. Highest score in Sensor A lab with an overlapping solving style was 98%. (Order Staying sequence before leaving sequence)
Sensor B used direct overlap but with staying input before leaving input. None of the fold changes were impressive. But this may also account to the strong input B both for binding too strongly when first attached, but also likely being way too fond of binding with itself thus taking a long time to bond. Something that shows up as high KD on for the B input in particular.
The majority of the Sensor C winners used a partial overlap with the order that I prefer. Leaving input before staying input.
So while we still have very few winners for RIRO labs, judging from the amount of winners and fold change, it looks like there is a preferred order.
What could we use preferred input order for?
If this trend with better fold changes for input/reporter overlaps with leaving sequence before staying sequence continue as I expect, then we can play this to our advantage.
To simply know when we have inputs that can not be made overlap in preferred order, then we should instead aim for strategies, like intersecting inputs or spreading out the inputs in the sequence, to raise our chances of hitting a better fold change.
Intersecting style is starting to show great promise too. Especially with the reporter packed away in the middle of the input. This type of exclusion strategy may end up getting us even further than a more direct overlapping exclusion.
But for now it looks like simple labs with two states, 1 reporter and 1 input, where reporter is on in one state and input in another, that these exclusion type labs benefit from a specific order of overlap. They are still among winners, even if overlapping opposite way. As they have been before too.
So if reporter and sequence are a good fit in preferred order overlap, they will be easy to make work. But if they don’t, switch strategy.
Overlapping style designs with direct overlap, with staying input before leaving input. (non-preferred)
Direct overlapping designs, with staying input before leaving input (non-preferred)
Direct overlapping designs, with leaving input before staying input (preferred)
Now I have some input behind the why. It has to do with KD.
Put leaving input before in order than staying input
I started to wonder if the order of inputs would affect KDON. I checked and the answer is yes.
While I first observed the benefit of putting the leaving input before the staying put in designs where these sequences overlapped, there even seems to be a preferred order even when the inputs are not overlapping.
The order of the inputs affect KDON
Leaving input before staying input → lower KDON
Staying input before leaving input → higher KDON
Clearest I can show this difference in the A/B labs with the inputs in predetermined order.
Leaving input before staying, general lower KD, lots of winners
Leaving output before staying, general higher KDON, fewer winners
If KDON is 15 or below things are fine. This means the design gets a top points for the baseline score (Which is max of 30 points of the total score).
The A/B lab with alternative predefined binding which has the fewest winners of the two, also has the highest KDON. Most of the winners are not getting full baseline score.
Yellow stars marks the designs that do not use direct sequence overlap and that are therefore not relevant for what I’m interested in.
First the designs sorted after fold change. Notice that Sensor C lab with preferred ordered input, pops out at top.
NB, I got the Ro lab somehow smuggled in my data search below. But since it’s highest scores are in the 80’ies, none of the design will pop up since my score limit was 94%. So ignore the RO bit. :)
Now I sorted the designs after KDON. What stands out here is that the Sensor C designs (preferred order) have generally lower KDON compared to both Sensor B and A. (non preferred order)
Now the Sensor B lab are not any anywhere near the critical KDON 15 limit, Sensor A is. It was possible making winners in all labs, although not equally easy. But for labs that have their KDON close to the limit, having the input in an order that do not additionally raise KDON, could be critical.
For the Sensor A designs like Run 1 with really high fold change, they have an intersecting solve style, where the reporter is hidden in between, two halves of the input. I expect this style to grow strong for round 2. This design also have a low starting KDON. (1.11 - whereas many of its winning peers had way higher KDON) So there are different ways of lowering the KDON so it gets in a range that is suited for what input one has got and to get beyond its inbuilt limitations, like non-preferred order.
I think we can use the KDON as a predictor of which designs are worth modifying to make winners faster. It allows us to get a preview of which inputs to put in what order. And where there is room for improvement. Even more - it can help us tease out strategies to get designs that are close to what we want, in the wanted KDON range for success.
Like the Sensor A designs that will trend toward having a too high KDON, if solved with sequence overlap in non preferred order, but are more likely to get staggering success if solved as intersecting style, as this allows for a far lower starting KDON.
Riboswitches - it’s all about the symmetry...
When symmetry is bad for static RNA designs as it causes misfolds, the logical conclusion follows that symmetry is good for riboswitches that needs to get moving...
Image taken from an earlier Riboswitch on a chip analysis post. Notice that the top left image which is close to a mirror image of itself, also has the biggest fold change.
While the ON state seems particularly fond of symmetry, I even think that if one makes symmetry in both states, there will be double bonus.
(Image recycled from the Same state 2 which is the most successful Riboswitch on a chip lab.)
We can use that riboswitches seem to dig symmetry or near symmetry, to take better aim at the target.
Omei has made several observations that I consider crucial for making better lab designs. I have written an intro to his ideas plus to the switch graph. Spvincent has been making good questions again. :) I haven’t got them all covered. The intro is a work in progress.
Here are the main points I will cover:
1) Coaxial stacking is helping reporter binding. Omei has observed NUPACK wrongly scores a design less favorable when the reporter is right next to one of the inputs. But he suspected that this would be more beneficial and made a lab experiment to demonstrate.
The designs that had the reporter closest to the input got a lower and better KDON. Coaxial stacking gives the reporter a better chance attaching at low KDON, something that will raise fold change potential.)
2) The fluorescent tag is affecting our experiments.
Omei is hypothesizing that the fluorescent tag on the reporter is influencing our experiments.
The fluorescent tag is attached at one end of the reporter. What Omei hypothesize is that the difference there was in his experiment with the reporter having either its 5' complement or it's 3' end next to an input, is due to the fluorescent gets in the way of optimal stacking.
We didn't knew which end of the reporter carried the fluorescent tag. But we decided to take a stab at guessing which end the tag was placed.
Omei placed our bet with Johan. And we won! 5' end it is. :)
3) The fluorescent tag is affecting input order.
Omei also thinks that the fluorescent tag positioning could be responsible for the pattern I have observed, that one input takes preferred order compared to another. As mentioned in the post Prediction for RIRO labs.
The order I claim is the preferable with leaving input before staying input, is exactly the one that buried the area with the fluorescent tag deep.
I added position of the fluorescent tag in on parts of images from the above mentioned post to illustrate.
Sensor A RIRO example with staying input before leaving input (non preferred order) + tag (not buried)
Sensor C RIRO example with leaving input before staying input (preferred order) + tag (buried)
4) Reporter binding is influenced by base sequence after the reporter.
I noticed something funny. A good deal of the best designs have either an A or a G after the reporter complement where the fluorescent tag is attached...GAACUUAg or ...GAACUUAa. (Big bulky purines - guessing perhaps via their size or electronegativity having a role in pushing the fluorescent tag off while not strong enough to prevent it from binding if it has access to the rest of the reporter). If one instead searches with U or C at this position there are far less designs doing well, neither on score or fold change. It even appears this work if slided a base. So either two AA's after the reporter input or two G's. Even a G two bases away seems to have a similar effect, sometimes even if there is a C before.
There even seem to be a similar pattern too at the 5' end of the reporter input, with A in particular but also G doing well before the reporter input, but that bit is not yet as clear.
I noticed this pattern, after I did a pull of the designs in the TB round 1 lab that had the lowest KDON, among which a bunch of winning acinc designs turned up, that had their reporter right next to the input (supporting evidence for Omei's coaxial stacking idea).
Flush stacking - reporter next to input
G base at 85 after the reporter.
I also noticed that these same low kdon acinc winners with best fold change had a G after the reporter input. Second best fold change had an A. That got me curious and I started to look into if there was a pattern. Which there were.
Omei added supporting evidence for my idea:
"Last night I observed that a series of A AND B designs had an inexplicable (by me) high KD_ON value. And what the highest ones seemed to have in common was a C just after the reporter complement."
I hadn't identified the C and U's as higher KDON's, just as worse scores and fold change.
Advice for this round
1) Place the reporter in such an order that you can quench (kill!) the end carrying the fluorescent tag, when it needs to be turned off. This can be done either by direct overlap with an input, sequence targeting this reporter end for turnoff or by a base sequence after the input with a calming effect.
2) Drag the reporter closer to an input to allow it to coaxial stack.
3) Also I will love to see loads of experiments with all possible base combinations 1-2 bases after the reporter complement in good and close to good round 1 designs. (And similar for the base area right before the reporter complement). So we can better understand what effect the fluorescent tag has on our results.
As Omei says: "We should focus on understanding what affect the Cy3 really has. So maximizing the score for the Cy3 may or may not be a good strategy. But understanding how big an effect it has will be important knowledge for making progress, regardless.
While I was watching the first round ABC2DEC lab results, it occurred to me that I could probably see 3D imagery of it if I tried.
Here is why I found it worth trying.
The best scoring designs are fusion style solves. A solve type that Atanas came up with, where one half of the puzzle is the smaller B/C puzzle part and the other half is the smaller A/C puzzle part. These two halfs are very similar in solve pattern, meaning they are alike so much as they can mostly be overlapped. (Demonstrated how to fuse two such puzzles here).
Now the B/C and A/C part don’t have their inputs in exactly the same order, while they are close. The B/C part uses intersection which the A/C part doesn’t.
I did a raw view of the top scoring designs and there definitely is some 3D stereo effect.
However to give it a better shot, where the designs are lined up vertically, I did a sorting after the top scoring design. This gave a better 3D view.
So I can only imagine that a version where all elements in the half parts are in same order, should create an even better 3D image. :)
A stereogram will only happen if there are repeat elements and sequence.
Quote from WIKI:
“When the brain is presented with a repeating pattern like wallpaper, it has difficulty matching the two eyes' views accurately. By looking at a horizontally repeating pattern, but converging the two eyes at a point behind the pattern, it is possible to trick the brain into matching one element of the pattern, as seen by the left eye, with another (similar looking) element, beside the first, as seen by the right eye. With the typical wall-eyed viewing, this gives the illusion of a plane bearing the same pattern but located behind the real wall. The distance at which this plane lies behind the wall depends only on the spacing between identical elements.”
I wondered how the data of the high fold change double aptamer riboswitch by PWKR that I had accused of near symmetry (in the above post) would look.
Double aptamer with stereo effect
And it results in an even stronger 3D stereo image. :)
Now I wonder, can bots see stereo images? Basically I wonder if one can get them to search for potential riboswitches in data. :)
Which made me wonder if one can also see stereo images in classic eterna data. And one can.
At least if one pick one of the more symmetric puzzles like the Star. And it is pretty easy in the Branches, as this lab basically is all repeats and symmetry. The closer the repeats are to each other, the easier they are to make a stereo image.
Even the grid and small letters repeated horizontally should be enough to create some stereo effect. But when letter sequence are repeated, it will create color overlap too, which will enhance the stereo effect.
I can't help thinking when looking at that Cy3 dye in the paper from Johan that Omei mentions, that it looks like something with potential for being a wire. Like resonance structure as in a conductive polymer.
Short resonance structure. Electrons moving around so the double bonds shift position.
Plus DNA wires are possible. :D Which is wicked cool and something I had been wondering about - okay more about if RNA could be conductive. Anyway - close cousins. :) Here is the paper I dug up.
Now as the Cy3 tag can floures, a resonance structure and a running stream of electrons in the dye would make sense.
However I think this go further than just a small localized wire. I can’t imagine the Cy3 and reporter bound against the RNA, without thinking about an electric circuit.
The first and foremost resonance structure I learned about is the benzene ring. Basically it is a very stable 6 carbon ring, with delocalized electrons whizzing around.
There is something else containing aromatic rings. RNA bases have them, just as DNA bases.
The Cy3 has two aromatic rings and these rings can be stacked on top of the RNA bases. And from what I get from the paper from Johan:
“NMR structural experiments established that both Cy3 and Cy5 dyes are mostly stack onto the terminal base pair, in a similar orientation to an additional base pair  and .”
So basically Cy3 could stack it’s “base-pair” on top of the RNA basepairs and become part of the temporary stem formed by the RNA and the reporter binding.
I’m aware that RNA is way more unstable than DNA and as such current will have a worse chance of transferring over a longer distance. But I think short distances could be enough to enhance fluorescence and/or affect RNA folding.
Basically I think there is an interaction between the aromatic rings that involves electrons. One thing I took note of in the paper:
"Typically, annealing of the probe to its target alters the fluorescence properties of the cyanine-labeled oligonucleotide."
This I read as a potential circuit may get electrons from somewhere else or lose electrons to somewhere else.
Here the bases are seen from a side view.
I have no idea which way the electrons could run. But I hope you get the idea.
An electron can jump from the electron sky around an aromatic ring to the next, if they are stacked on top of each other. So I basically see the fluorescent tag as closing the circuit.
Which lead me to another thing I have been wondering about. If an electron could pass through a hydrogen bond. Also I started thinking about quantum tunneling. Since there is no other connection at the other end of the Cy3.
I read a very interesting paper, that says that just that can happen. Electrons do prefer to jump between rings stacked on top of each other, but they can jump too between aromatic rings in different strands. (Fig 3.)
“Thus, ET (electron transfer) proceeds preferentially down one strand in double-helical DNA.”... ...If H-bonded basepairs must be traversed the ET kinetics slow considerably.
Electron Transfer Between Bases in Double Helical DNA (Sorry, paywalled)
Electron donor and acceptor in the DNA base world
I learned one more hilarious thing. Even DNA bases can flouress - or rather slightly modified versions of Adenine. :D
These artificial fluorescent Adenines fluorescence can be quenched by a Guanine and another fake base.
“The fluorescence of both of these bases is efficiently quenched by deazaguanine (Z) and guanine(G), with only small amounts of quenching observed with inosine(I) and the other DNA bases with significantly higher redox potentials.”
(From above mentioned paper)
Guanine and Adenine (mutants) can behave as electron donor and electron acceptor.
Guanine - donor acceptor
Adenine - electron donor
(From above mentioned paper)
And when I look at the G behavior after the reporter complementary site, I very much see quenching behavior. That’s what I seen from the start, I just mainly attributed it to Guanine size and it taking the spot so the tag couldn’t get a good grab at this side of the strand.
Will be interesting to see if the Guanine quenching trend after reporter compliment continues.
Guanine as quencher
Perhaps this could be reason why GU’s and mismatches are less welcome inside the reporter stretch is that not only do they prevent the reporter from binding well, but also the geometry disturbance prevent the electrons from flowing.
“Moreover, a profound sensitivity to stacking has been observed; in the present of base mismatches or other stacking perturbations, long-range ET (electron transfer) is essentially turned off.” (From above mentioned paper)
In that case it would make sense why even input complement sites can be affected too if they are close to or next to the reporter.
Since G can be a quencher (electron acceptor) in DNA, I wonder if it can also be an electron donor?
Reporter bases, the complementary bases, neighboring bases + even further away bases, affect glow
I think this circuit is affected by base sequence. The paper Johan shared, says something about sequence affecting stability.
The paper says” The magnitude of stabilization depends on the base sequence”
There is also a geometry angle to it. The aromatic rings are stacked in RNA - they are lined up for electrons jumping through the aromatic rings. Or almost. Purine and pyrimidine bases don’t have the same size.
Top drawing I have tried illustrate the size difference between bases.
Bottom drawing: I suspect electron transfer would work better in particular if the same kind of base is on top of each other. Like purine on top of purine or pyrimidine on top of pyrimidine. And even more so if the same kind of base is stacked on top of itself. Like in poly(A) or Poly(X)
Here is a paper that seems to be saying something similar about DNA.
The common model for electron transfer through DNA is based on overlap between p orbitals in adjacent base pairs. Irregular base-pair sequences may lead to localization of charge carriers and reduce the transfer rate of electrons[9,15]. A structure containing a single type of base pair may therefore furnish the best conditions for p overlap.
Direct measurement of electrical transport through DNA molecules
I think the base type difference between the reporter could create a potential differential. Just voltage difference is needed in an circuit to get current flowing.
I think our lab results will differ just depending on if the reporter mainly has a purine bias, a pyrimidine bias or a more balanced base distribution. I strongly suspect this will affect the reporter binding and KDON values. I don’t think that every reporter will be equally good. I simply think that the individual kinds of bases affect the electric conductivity and circuit that I strongly suspect occurs, when the reporter and the Cy3 fluorescent tag together bind up with our RNA design.
What I think is that the reporter or part of it along with the tag functions as a circuit and this somehow affect how well the reporter will bind and how much it will glow.
Even more extremely if the reporter itself and its landing site were pure poly(X)es.
I think cytosine and uracil are bad conductors, since they are smaller and more unstable - wiggly.
However I think guanine and adenine are better, since bigger bases are more rigid.
Poly(A) and super positioned electrons?
This thing with electron flow may potentially have implications for the strange poly(A) pattern that turned up in some of our classic EteRNA labs.
And since I have recently got a crush on quantum mechanics, I can't help see quantum tunneling and a standing wave of electrons, through the poly (A) stretch. Perhaps even the weird break in pattern could be reflection.
However the weird poly(A) pattern dies if interrupted by any other bases than A. (Any base change would affect local geometry)
The poly(A) structure is different in that it do not seem to be part of what seems a circuit. There is only a single stranded RNA.
My sum up of what is needed for polymer conductivity based on the book: Polymer chemistry - An introduction, Third edition, page 117-118
Delocalized electrons. For backbone to get conductive a larger range electrons
Doping - something that in chemical world can be done with metal.
Geometry matters. The more rigid, the more crystalline
In other words stacked aromatic rings could allow electrons to jump from ring to ring + some quantum tunneling (quantum coherence) should an interstrand jump be needed - and thus enhance glow.
Our RNA’s bathe in MG2+
I suspect Poly(A) will do very well on being crystalline.
Did you notice any designs for A*B/C*C that looked promising but give poor scores? Or vice versa? Even 1 or 2 designs would help us.
Quantum entanglement in partner labs
I have been making sum ups of different switch labs through time, when I found a pattern that I thought applied. However I have long been having a feeling that there was a more general pattern, across whichever reporter and input was used.
I already mentioned that for the logic gate labs, it seemed as if each lab had its reverse partner. Like the AND and NAND labs. The AND winners would have their RNA inputs exactly opposite in order to those in NAND. (See examples at bottom of the post)
Partly due to that our new reporter allow for a lot more variance. Also I had trouble digging it out due to confusion about what was an ON and a OFF switch.
What is an ON switch, what is an OFF switch?
I have ended agreeing with myself about that an ON switch is all switches which have its Reporter binding in the last state. Similar an OFF switch is a switch which has the reporter not bind in its last state. No matter how many ON or OFF states there are besides.
Partner labs and input reversal
The idea is that if you know the favored input order of one lab - you know the favored order for the partner lab too. Even just knowing what type of lab it is, already helps. If you know if you are solving a single input OFF switch, you also already know something about what it likes - like getting the input placed at 5’ end.
Usually ON switches score better than OFF switches.
Omei have noted that the OFF switches tended to have a higher KDON than ON switches.
I have highlighted (yellow) a few cases that did not follow the general pattern. The two first, I suspect being due to the strong input being really strong and as such not care as much about where it lands, compared to a weaker input. The last highlight I have no idea why.
I have looked at the main trends inside a lab - meaning it is possible finding winners that follow a different pattern, I have just cut out the main trend.
Partner labs - Example images from Logic gates lab
Notice the reversed inputs between the labs. Both designs are winners. And they each do the opposite RNA math of the other.
OFF switch, MS2 is off in state 4 when both inputs bind
ON switches, MS2 is on in state 4 when both inputs bind.
Here are some things I think will work and help us get better designs. I think Input order, symmetry, structure repeats and state repeats will be key.
1) Input order
Have the strong B input before the weak A input.
- The ABC2DEC highscorers seem to prefer this order and so do the simpler ABRO variant. (OFF labs)
Abro variant, score 80%
Have the weak A input before the strong B input.
While it is currently not the case, I expect the comming ABC2INC high scorers to have the opposite ordering of the A and B inputs of the ABC2DEC lab and follow the trend that pops up in the simpler ABRO variant. (ON labs) The ABC2INC design that Omei picked out as doing well on all concentrations, follow this order.
Abri variant, second highest score 80%
(More about input order)
Many of the highest scoring designs for round 1 of the TB lab show a strong dash of symmetry. (Particularly for state 4, but also state 3.) The same is the case for a good bunch of winners from earlier switch labs. ON switches tend to show stronger symmetry than OFF switches.
(More about symmetry)
3) Repeat states
Spvincent asked me a question in relation to why I in particular liked one of his hard DEC lab puzzles. (I like the whole series) As for the discussion to benefit all, I bring it up here.
Here is what he said: "Tx Eli but I'm intrigued why this design might be promising. For these 4-state A*B/C*C puzzles I've been trying to arrange things so that each state has a different secondary structure (is that the right term for RNA?). It seems natural and more appealing aesthetically. Yet here, as is the case for many other published solutions, the first 3 structures are identical."
Yup, you got it exactly right with secondary structure.
It is also great that you create a huge variation in your designs. Spreading your investment.
Actually state 1-3 being the same is one more thing I like about your design. Here is why.
Already in the logic gates lab designs popped up that had two or two and two states being the same. And they were among the winners and highscorers. Think about it like this - if two states can share the same geometric structure - there are simply less things that can misfold. I think it allows for a more precise switch. Fewer things that can go wrong.
Following that logic, the more states that are able to share structure, the better.
What I particularly like about your design is that it is close to symmetry and especially that the subpuzzle parts are repeats of each other - both are things that would be sins if in excess in static designs - but I have come to regard it as a stamp of quality in switches.
4) Structure repeats
I'm not just seeing symmetry in switch winners or highscorers. I see structure repeats. Like two sub labs fused for a hard design, and having the inputs in the first half of the fusion come in same order as in the second half.
Spvincents design mostly follows the pattern of first rounds best DEC fusion designs. (cBr(B) cAr) By repeat sub parts, just with a different order. (Bcr Acr)
However Spvincents series has the advantage that it has the inputs in the BC part in the same order as the Whbob design that is the best ACDEC design that can work for a fusion and got a highscore (93%) in the first round. While what works for a sub lab may not necessarily be good for a fusion in a hard lab, there still is quite a good shot that it will actually work.
I have drawn an overview of the puzzle. (cARBc) Notice the symmetry and mirroring. (The puzzle itself isn't fully symmetric, but is near symmetric in state 1+3 (Ignoring the static stem).
I did a guess (rcBAcr) for what the ABC2DEC might like based on the INC one. Here I copy the mirroring part. This design doesn't have sublab repeats as the Spvincent design I mention in the post above.
Also the bcdec part does not have its inputs in the optimal order as for a solve in the bcdec sublab, however I still count this worth trying as it may work for the hard puzzle on overall. Just as the design with its concentration that goes in the right direction, doesn't carry full subparts of two smaller sublab designs, but only of one. The two sub part puzzles are sharing reporter. Both doesn't have one.
Help me fix the promising ACB2INC design
This hard INC design Omei picked out, had really high KDON (56 - and it should preferably be under 15). So I have been thinking about how to lower it.
I decided to pull the trick from the Riboswitch on a chip lab where the absolute winner of this lab, used a double aptamer and came out as the design with the highest fold change.
So I have simply doubled and even trippled the reporter binding site, hoping that more reporters will bind in state 4 so it will ease the reporter binding and thus lower KDON. I also did some modifications of the original that I think may improve it.
I will put up some small design series of these types over at the forum post Promising sequences to mutate for this lab round.
You can help me trying to fix the problems with this ABC2INC design.
I noticed that ON switches tends to be needing fewer bases for a solve than OFF switches. ON switches often carry an extra static stem or two. Where OFF switches trends towards using the full space of the sequence length.
These are the labs where I see the trend.
Here are two partner lab examples
XNOR - ON switch - Long static stem for hiding away bases
XOR - OFF switch - use of full sequence
This with ON switches being shorter than OFF switches, in particular happens in puzzles with two or more inputs.
The length pattern in relation to ON/OFF switches, is not universal for all switch labs, eg, some exceptions are the R2 and R3 lab in the A/B labs. They show reversal of the pattern. But I find it interesting that the general pattern is so strong already. Keep an eye out for this. It could help us pick the right solving style to go with the lab type.