Switch Puzzle with Psuedoknot loading issue

I recently created a switch puzzle that contained a pseudoknot in the first state, however it will only load in the two state switch puzzle maker, but will not load as an actual puzzle. I believe this is the first switch puzzle containing a pseudoknot. Is it possible to fix?


Link to puzzle: Eterna

Hey TI.

This is a bug - if you enable pseudoknots in one state, it should enable them in both. In typical switch puzzles, you’re actually changing the environment of the RNA somehow (eg, adding a small molecule to bind to), and saying “I want an RNA that works in both environments” - in this case though, you’re just telling the folding engine to run its predictions differently. As far as I know, that isn’t really a valid scenario.

Unless you (or our scientific staff!) have any objections, I’ll unpublish the puzzle.

LFP6 by “bug” I presume you mean a bug in the puzzle maker, in that it is allowing different folding engines to be specified for different states.

I wouldn’t expect there to be any problem with a switch puzzle where some states fold into a pseudoknot and some don’t.

Yes, thats what I meant

You may go ahead and unpublish it if there are no objections from the scientific staff.

okay, so the problem is that I’m not actually changing the enviroment of the RNA except for having the pseudoknot in the target structure for one of the states? A little earlier today Lroppy was saying it could be naturally possible for RNA to fold this way, but the folding engine may or may not fold it differently. Is this completely unnatural and the folding engine just has a bug within it? Or could it be possible to fold this way but the folding engine doesn’t recognize it outside of puzzlemaker?

Also if I republish it, except I add a protein binding to the second state, will it work? As Omei was saying the probem isn’t with the pseudoknot right?


I tried one of mine and it sort of looks like the molecule makes state 2 go back to vienna instead of staying with NUPACK. I can’t publish until tomorrow due to 3 puzzle limitations. So I will update then.

By the way when I was making this puzzle I considered using the molecule in the second state (the one without the psuedoknot).

I republished the puzzle with a protein binding in state 2, but it still is not working, so are you sure the problem is that there are no enviromental changes?

So the bug is allowing different folding engines in the two states. Doesn’t have anything to do with pseudoknots or NuPACK per se. The issue is coming up because we now are working with a third type of dot-bracket notation and we didn’t previously.

Yeah I wasn’t completely clear, sorry. I meant that it shouldn’t be possible for you to ask NUPACK to consider pseudoknots as possible in one state, but not consider them in a second state - if you want NUPACK to consider pseudoknots as something it should include in its set of potentially valid structures, it should do it consistently. Adding a molecule actually wouldn’t help the problem, as NUPACK (our at least our implementation of it) does not support both pseudoknots and molecules.

Consider what you’d be asking for - “I want a sequence that both folds into this structure and doesn’t fold into this structure at the same time, without changing anything about the RNA”. The only difference is how you come up with the prediction, not anything about the thing you’re trying to predict - and I don’t really see any real meaningful information that would give you (vs, say, folding in both NUPACK and Vienna - it could be interesting to find something where the same structure is satisfied in both models, which is something I eventually would like to support).

It would be theoretically valid (if it were supported) to say “give me a sequence that folds into this structure which happens not to have pknots without a molecule and this structure which happens to have a pknot with the molecule” - but you would tell nupack to use pseudoknot prediction in both states, and your goal in state one is just that NUPACK shouldn’t predict a structure that has a pseudoknot with that sequence

The confusion here is probably the lack of clarity of what you’re seeing in puzzlemaker - one state is calling nupack with a parameter saying “pseudoknots are a thing which you should consider” and the other is saying “eh, don’t worry about whether or not there might be pknots, because that’s slow”.

Oh ok I see now, that makes sense, thanks!

I have one other question, is this an issue will be fixed in the near future? Or will it require more time before we are able to create pseudoknot NuPACK switches?

Frankly, I don’t even know if nupack supports molecules and pseudoknots at the same time. It doesn’t support multistrand with pseudoknots. The only other type of switch I could think of off hand would be one where we change the temperature, but that’s not something we’d be able to support until the rewrite. Regardless it’s not part of any immediate plans.

gotcha, so at least for now i probably shouldn’t attempt to make any more switches with a pseudoknot and a molecule.