EternaFoldThreshKnot Puzzles Is it more than a guessing game?

Am I the only player having problems with EternaFoldThreshKnot puzzles?

Learning something new is always difficult and that is one of the challanges. We players would not be here playing and making the puzzles along with taking part in labs if we didn’t want some type of brain challenge. But EFTK is somewhat different.

Solving these puzzles seems more of a guessing game than any learning or intuituion. Also, there is no free energy guide to assist players. We have a small tutorial series which offered some assistance thanks to sgeldhof.

As a player, I can carry on player the Vienna, Efold etc., puzzles that is no problem, but not very rewarding. What I do see as a problem is that as new players join Eterna, they see the newest puzzles at the top of the screen automatically and these are probably the first they to choose to play. Therefore, having the most difficult puzzles, which have only a few solvers as the first that the new players are offered is unlikely to encourage people to stay.

As the EFTK puzzles often have less than 10 solvers, many have 5 or less solvers, this indicates these are difficult (which in itself is not an issue), but failing to solve again and again maybe off-putting if you have only just completed the tutorials.

I would just like to understand how to model the Engine better, and encourage new players to stay, rather than discouraging new players and those of us who do not understand EFTK? Also, would a mix of EFTK/EFold puzzles be possible?

I’m just putting this out on the forum as a couple of people have suggested that they are also having some EFTK issues.

1 Like

Thanks jenni11 for bringing this up here in the forum, I agree with this. A solution could be threshknot new puzzles would not come up unless filtered in.

1 Like

Thanks Jenni. I have the same feeling of continually failing and the few times I succeed I am sure it is by chance. I have to say that this does not mean that these puzzles are not interesting since a subtle play of forces is involved, and it is interesting how balance is gained or lost due to a small mutation. But, as it is a game in which the reward encourages you to continue playing, I am not sure that new players will be disappointed.
Surely we will find some way to make these puzzles more rewarding for the player. I can’t think of anything at the moment, but there are good brains with great ideas here.


As far as the solving process of EFTK puzzles: As I’ve mentioned elsewhere, you should be able to get a lot of useful information and guidance by looking at the dot plot. The core way EFTK works is by picking out the pairs that have the highest likelihood which you can see pretty directly via the dot plot. Ideally in the future we can make this more usable by eg visualizing that data directly on the 2D strand.

With respect to the difficulty being offputting, these puzzles being more visible, etc: What you see first on the puzzles page is whatever’s newest. We can’t (and shouldn’t!) change what is surfaced there. We have some general ideas for other ways of browsing puzzles (which would be more of the default/first thing seen particularly for new players) as well as improving rewards and general feeling of progress (which has been a discussion for a long time!).

As with many things, we unfortunately haven’t had the capacity to prioritize this among everything else on our plate.

1 Like

Using the Dot Plot to make (a lot of) lab puzzles in the OpenKnot Labs taught me to respect the balance of energies between the knot and the rest of the puzzle. It’s basically a energy free-for-all. First you have to find the segment of the puzzle competing with the knot, then you have to balance those energies so that both exist at the same time, all the while making sure adjacent structure do not interfere with everybody else and mess thing up. Master the technique and you to can become a thermodynamic RNA magician.
I created the difficult ones by making energy that doesn’t really exist.
The “junk” in the loops is not “junk” in a EFTK puzzle but is a integral part of the mess. I’ll make some easy ones next week for you to practice with but you won’t learn much without using the Dot Plot to solve. PS: The difficult ones I have been making are to annoy Astromon and AndrewKae and also to challenge myself while investigating EFTK…


Well Jenni, you really struck a nerve. I have to talk myself out of crying when trying to solve these puzzles!! Only half kidding. I wish I could use the Dot Plot but I don’t understand it. Is there information out there that explains it? I think I need a Dot Plot for Dummies manual.


P.S, Despite my ineptitude in solving these pknot puzzles, I do appreciate having them and the skill it must take to make them. I gave it a little bit of a try to make one but no success.

1 Like

Thanks for your description, and I would appreciate some practice puzzles.
You are doing a good job of keeping Astro and AK on their toes :wink:

mjt, I am with you, feeling quite dispirited about this. But hopefully as we are talking about this now, we might be able to get small steps towards improving our understanding :+1:

1 Like

I’m still leaning towards we have found a way to exploit the model that will prove to be counterproductive. I have been building these structures of unbonded pairs to solve some of these, and it is exactly like building a switch. Seems like getting rewarded for slippery solutions.

1 Like

In the mean time solve some simple switches, same balance the energies problem. On RHS select “most solved” then below select “uncleared” then at “states” select 2 and 2. This gives you a list of simple switches to work on…
I make EFTK puzzles either by copy/modify method or copy a structure I want to work on into a text editor and manually add the brackets where I want them, then replace the structure back into puzzle maker…
You know about “beam to puzzle maker”, right click a puzzle and select from dropdown if not.

The irony of this is, I made a EFTK, ‘give this a try’ locked some of the bases (on one side of the pk) so that I thought it would make the puzzle easier to solve, and I still cant solve it myself.

I even made the puz with more locked bases - still no joy. I know I could cheat, but that really is not the point.

I see that there are several solvers in the series of those puz. After 24 hours, I am going to add another to the list with even more bases locked, for those of us that need a helping hand.

The upper right triangle in the dot plot displays the pairs currently being predicted in the ensemble with the most frequently predicted pairs in the darkest color. The bottom left triangle is empty for these puzzles so just ignore it. Hovering over the dots displays the base numbers of the pair. It takes a few tries for the brain to get accustomed to the alternate representation style.

Here is the Wiki page. The directions for detecting a pseudoknot are different than the Wiki page discussing a classic Eterna structure. I believe what JR does is look for the second pseudoknot stem in the dot plot along with the first stem.

Here is the dot plot of my solution of Baby’s First Pseudoknot.
Screen Shot 2023-09-23 at 6.36.50 AM
The first stem runs from 3/40 to 14/29. The second stem runs from 20/49 to 23/46.

I unsolved the puzzle and the dot plot no longer shows the second stem.

Screen Shot 2023-09-23 at 6.45.33 AM


Thank you, DigitalEmbrace, This is helpful. I thought one dot represented one nucleotide, so now I know it represents a pair, and I mistakenly thought the lower triangle played some part, so you cleared that up for me too. The phrase "clean up the dot plot " I don’t understand or what all those other light dots signify, and also it seems like it just shows what one has done in the puzzle. I still don’t get how it helps in solving or making the design better. Maybe now when I try using it, I’ll understand more. Thanks again.

That’s what I do. Look at the second chart and you will see only the stack is really black meaning 100% pairing probability, that’s it ,nothing else is going to happen. But look at the first chart circled in red, the stack is grayer and the knot (small circle) shows up, meaning to me the pairing probability has been distributed from the stack to one that is now shared with the knot. Compare backgrounds and you will see the bottom clear and the top with shadows. The shadows are pairing that you don’t want but may not be able to get rid of (you want to minimize those background pairing) . The pairing @ 31/E is real and could mess up the design when tested. Get rid of the background pairings if you can. The ideal design would have the stack (big circle), the knot (small circle) both same color gray (as dark as possible) with a clear background. This ideal design would work(and has) because there is nothing else for the design to do except flip between the knot and stack giving a high scoring design…


Let’s interpret the 2 charts above only using pairing probability (PP) language. Think of thermodynamics as taking photos of the design over a specific time frame, then adding these up into a graph, that you can then read, telling up the success of that sequence to conform to your design. You mentally add up all the diamonds shades to = 100%. So the second chart would be about 97% stack and 3% background. The first chart looks to me 80% stack, 6% knot, 5% stack @ 31E and 9% background. I would want my knot stack to be a higher % and I can do that by reducing the energy in the large stack and stealing as much energy from the 31E stack and background, which results in putting the energy back into the knot stack, increasing the knot’s PP at the expense of the other pairings, aiming for 50% PP stack, 50% PP knot and 0% PP background, . Flip, flip, flip but with a purpose.

1 Like

How does this apply to solving EFTK player puzzles? Do you look for areas that are dirty on the dot plot and try to block the unwanted pairings?

Something else that hasn’t been mentioned is the dot plot is essentially a different visual representation of the pairing probabilities displayed in the ArcKnot tool.

It doesn’t at all. It tells you when you have a good lab design, and since labs for me trump puzzle designs, that’s what is important. The CONCEPT helps because you have to create balanced energies with bogus pairs that help stabilize the puzzle. Sometimes a random G placed in the middle of nowhere solves the puzzle. EFTK puzzles are rather maddening. The loops and bulges now become a significant part of the puzzle.
Understanding the PP concept increases your patience and perseverance to eventually solve the puzzle.
Here is the solution to Double Trouble PPartial:

There is more stuff in the loops than in the pairs. Welcome to EFTK puzzles.

1 Like

I want to briefly clarify a couple things:

  1. A reminder/refresher to the way EFTK works: it reports a pair as “paired” if it both a) has a pairing probability greater than the cutoff (which we currently have configured as 0.15) and b) that pair has the highest probability for both partners in the pair. When looking at the dot plot, your goal is therefore to either increase the probability of that pair (by strengthening possible structures that include that pair) or decrease the probability of other pairs (either because some other conflicting pairs have higher probability OR because the probability of all other possible structures suppresses the probability of pairing you want below the threshold).
    If I had to bet, the very last part is what tends to trip people up. Why would putting a base in a seemingly unrelated position make an impact on some pair folding correctly? Because you’re affecting the probability of other structures in the ensemble, not just the target and natural structure. The ensemble is still an important when working with MFE-based models, but it is even more important with ThreshKnot because it uses the entire ensemble to predict pairs, not just the “best” structure in the ensemble.

  2. Per the bottom triangle of the dot plot being empty - this should show the target structure pairs (ie, as if the target structure were the only possible structure, with 100% likelihood). I believe this is actually a bug - I want to say EternaFold in general doesn’t properly return this part of the dot plot? I’d need to check.

1 Like

Most players (my guess here) march the Target/Natural total down to zero by whatever method they are used to which solves the puzzle. With EFTK puzzles that tool is missing which forces the player to rely on their puzzle intuition to solve the puzzle (sounds like some are a bit rusty). They have to learn the “jedi” thingy I guess. The Dot Plot concept is useful to learn.

1 Like