Area to discuss Switch puzzle analysis and its relation to lab

[Brourd]

I believe the tricks that result in manipulating the loops that a ‘molecule’ free energy bonus is applied to, should solve the above puzzle, given it changes the way the entire MFE structure is handled (from a puzzle’s point of view).

I think one possible solution to the dilemma of expanding the number of switching motifs in the puzzle is to increase the total number of puzzle states that exist. Currently, your puzzle has three independent domains that all switch between two different structures. Additional states would allow for the stabilization of these additional domains. Granted, the addition of multiple ‘molecule’ bonuses to each state would serve the same purpose, so there is always room for improvement of the game.

[Eli_Fisker]

MicroRNA based switches

Discussion between Omei and I from today. Added a few puzzle links as I made the puzzles later. 

Tonight I have been playing around starting from the question of what a switch would look like if based on a microRNA.
I took our old friend hsa-mir-208a and it sparked a pretty structure. I think it is perfect for creating a massive 15 state multistate puzzle. [Correction 16 states: https://www.eternagame.org/web/puzzle/8599451/ )

omei [12:05 AM]
I was thinking we used a short sequence from hsa-mir-208a as an input oligo. Have I forgotten something?

eli [12:06 AM]
That is correct

omei [12:07 AM]
For the current fun, are you using just that, or the whole mir?

eli [12:07 AM]

https://www.eternagame.org/web/puzzle/8599447/

eli [12:08 AM]
Just that. I tried one of the whole unmature microRNA and it was boring.
Most of the mir’s I tried so far are boring - not sparking much structure.
Our mir was special in that it has 5 A’s. The A’s seems to help spark extra structure. [Repeat A’s and U’s seems to help with that by sparking loop structure in Vienna]
But what made me really excited now, is a whole other microRNA.

eli [12:09 AM]

https://www.eternagame.org/web/puzzle/8599478/

eli [12:10 AM]
This one is perfect for a massive switch.
I found it in a paper on alzheimers and microRNA’s that are upregulated in alzheimers.
I think it is pretty. I wonder if microRNA’s can pair with themselves.
I mean not individually.

omei [12:15 AM]
When you use the word miRNA, are you thinking about the ~22 base segment, or the ~60 base hairpin it is cut from?  (the pre-miRNA.)

eli [12:16 AM]
22 base segment

omei [12:17 AM]
So our last picture is based on ~22 base segment that itself forms a hairpin?
Maybe not – I’m not sure why I thought that, looking at your picture.

eli [12:20 AM]
uploaded this image: Only when in NuPack

eli [12:20 AM]
But it harbours sequence that is complementary with itself if the sequence is in double.

omei [12:20 AM]
Ok, a relatively week hairpin

eli [12:21 AM]
uploaded this image: image.png

eli [12:21 AM]
I think my surprise stems from that I hadn’t expected anything beautiful forming from the microRNA’s. As they seem very repetitive.

omei [12:24 AM]
Repetitive in what sense?

eli [12:24 AM]
In that they are biased in sequence. Perhaps lacking one of the base letters.
Or very heavy in U’s

omei [12:25 AM]
I see.

eli [12:25 AM]
By the way U’s seems to be heavily used in full moving switches.
Example from Let-7 with no C’s:

ugagguaguugguuguauagu

omei [12:27 AM]
I suppose one “reason” for this is that they would be less likely to form dimers. (edited)

eli [12:27 AM]
Good point
I like. That could explain.
Also the sequence is cut from what is a hairpin stem and more specifically one strand of it. (edited)
And a strand in a hairpin is different in sequence from a full hairpin stem.

omei [12:29 AM]
But long single strands that form hairpins are also able to form dimers.

eli [12:29 AM]
Plus if the microRNA is too varied in sequence - it will be a good and very static stem. Less likely to detatch.

omei [12:30 AM]
… which makes me wonder if pre-mRNAs have anything, like chaperone proteins, to discourage dimers.
Looks like they do require one or more proteins in oder to escape from the nucleus.

eli [12:37 AM]
Is RNA dimers kind of like palindromic RNA

omei [12:37 AM]
Exportin5 looks like it would probably distinguish against the dimer form. https://www.ncbi.nlm.nih.gov/pubmed/14730017
ncbi.nlm.nih.gov
Exportin 5 is a RanGTP-dependent dsRNA-binding protein that mediates nuclear export of pre-miRNAs.  - PubMed - NCBI
RNA. 2004 Feb;10(2):185-91. Research Support, Non-U.S. Gov’t
Not sure what you’re including in palindromic.  Dimer itself means nothing more than two RNA binding to each other.

eli [12:39 AM]
Ok
Got it

omei [12:40 AM]
Any sequence that simply forms a hairpin can also easily form a dimer with another copy of itself.

eli [12:40 AM]
And since pre-miRNA are the same, they should in principle be able to bind with each other also. I think I get your point
And some miRNA’s were also downregulated in alzheimers. I may try make a puzzle of a dimirized preMRNA of one of those.

omei [12:44 AM]
It should work.
Remember in the the MS2 cooperativity puzzles, how the MS2 hairpin sequences tended to pair up against each other rather than forming a hairpin?
… in the absence of the aptamer boost.

eli [12:47 AM]
Yup.
It also seemed to matter what were in between them
In round 1 I placed stems between them. They were not happy.
In round two what worked was them being close together seperated by loop bases, and a weak stem at top.
So a loop is a very different seperator, compared to stems.

omei [12:49 AM]
In retrospect, I think the most cooperativeness might come from putting two directly adjacent.  We now know the energy model incorrectly estimates the energy of adjacent stacks, but we didn’t know that then.

eli [12:52 AM]
True. My jumping jack solves (in the RNA input labs) that had a stem between the input and reporter weren’t as effective either in the RNA input labs as the solves that had the reporter and the input fairly close to each other and forming a loop between them as they turned off. (edited)
The input and reporter also there benefitted from being close together (coaxial stacked) when both were turned on.

omei [12:55 AM]
As I recall, one of the highest, if not the highest, cooperativity score came with two MS2 hairpins separated by a single base.  We now know that’s close enough to get a small coaxial stacking bonus, but not nearly as much as having them adjacent.

eli [12:55 AM]
Ah, I didn’t realized that. Well noticed.

omei [12:56 AM]
Unfortunately, I don’t remember the specific design.  It was from a player who was not one of the “regulars”.

eli [12:56 AM]
I think I looked more of the scores.
As in EternaScores.

omei [12:56 AM]
Right.

eli [12:56 AM]
Yup. I recall there being a new player getting a real fine score
Christopher something

[cynwulf28]

There are a lot of microRNAs out there. I noticed this paper and wonder if it could help to make sense of the miRNA zoo we face: https://www.nature.com/articles/srep25941

[Eli_Fisker]

I love the image of an RNA zoo, especially that there is even one among the microRNA. Thx for the link. I shall read the paper.

[Eli_Fisker]

Same sequence - different engine - different structure

Chat with Omei from yesterday:

I have been making microRNA puzzles. I got a Vienna2 switch made. It made me realize that we need more switches done in Nupack and Vienna2.
The vienna2 switch I made from a Vienna switch. [Based on the same sequence]
They were quite different.

omei [8:01 PM]
So player puzzles are tied to a specific engine?  I figured that must be the case, but never verified it.
When another player tries to solve it, is the engine displayed?

eli [8:01 PM]
I did a small illustration with single states:

eli [8:02 PM]
uploaded this image: image.png

eli [8:02 PM]
Some puzzles are exactly the same despite engine.

omei [8:04 PM]
AFK for a few minutes

eli [8:04 PM]
Ok.
Same sequence, different engine:

eli [8:05 PM]
uploaded this image: image.png

eli [8:16 PM]
Cynwulf was hunting a microRNA fold and he switched engine to get a pretty structure. That’s what got me inspired.

eli [8:17 PM]
uploaded this image: image.png

eli [8:17 PM]
He ended up using a selfgenerated microRNA to make the above puzzles.
So I instead swapped a switch puzzle between engines.

Seing different structures of the same sequence

To see the different engines in action on the ground state in this puzzle

1. paste in this sequence:

AUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGUAUAAGACGAGCAAAAAGCUUGU

2. Right click and choose the option Beam to puzzlemaker

(Make sure you don’t have the puzzle in multistate view. That will make only one of the structures go to the puzzlemaker in double.)

3. The first state (left) is stable in vienna.

4. Click on the leaf for put the puzzle in Native mod.

5. Switch energy engine by clicking at the engine name where it says Vienna and there are a molecule, at the top left of the screen. Now you can see the alternative structures of the same sequence in different engines.

[cynwulf28]

A word of warning…when you switch your Engines you lose your ‘stacks’ and will NOT be able to use the undo arrow to a time before you last switched Engines. I have found that out the hard way more than once. 

[cynwulf28]

This might prove relevant: 

[cynwulf28]

https://www.youtube.com/watch?v=qZ9cTiqALdw

[Eli_Fisker]

Got one lab design made and a puzzle up
My current fascination is transposons, but I’m not done with prions either.
I can basically spark a switch as long as I make repeats - almost no matter what starting sequence I use.

omei [9:14 PM]
That answers the question I was about to ask – whether there seemed to be something unique about transposons or prion mRNA.

eli [9:14 PM]

Sometimes it takes a specific numbers of repeats to spark a structure, like 4 or 8
but some structures only happen if there are an uneven number or rather incomplete extra stretch of one of the starter sequences
I mean sometimes the pretty structure don’t turn up if there are 2 repeats
or 4

omei [9:18 PM]
Can you predict what you end up with before you see it?

eli [9:18 PM]
No.
Here is an example that when repeated in 4 is ugly. But if I add an incomplete fifth part, it gets pretty.
Sequence: AUGUUUGCUGGAUUCAAAGAGUUCAAAGUUUGGUCAAUACUUCUUGGAAUGUUUCCAGUAUUUGAGGAAA
CAUGUGACGAUGAUAACGACCUUUAA

(https://www.ncbi.nlm.nih.gov/nuccore/U63707.1)

eli [9:19 PM]
uploaded this image: image.png

eli [9:20 PM]
with a fifth incomplete sequence

eli [9:20 PM]
uploaded this image: image.png

(Made a puzzle of it afterwards: http://www.eternagame.org/web/puzzle/8663375/)

omei [9:21 PM]
Interesting.

eli [9:21 PM]
It happens a lot. Also with 2 repeats and an incomplete third one
So I usually try put in different amounts of repeats to see what they spark of shape
and try fill the whole puzzle also
I can likely make both puzzles switch, but the latter will be more likely to spark symetric switches
Pretty to pretty.
Its hard to go ugly to pretty
But very easy to go from pretty to ugly
Its kind of like entropy. More possible states = more entropy. If there is symmetric order, it is easy to spark more limited symmetric order. But if there is huge disorder, it is harder to go towards lower entropy
Repeat sequence is a way to make order and limit the amount of possible states

omei [9:25 PM]
Makes sense

eli [9:26 PM]
And I also think this is why successful lab switches are so fond of not just repeat sequence but GA and CU rich sequence
Plus likely some physical reasons

Afterthought

I made an earlier puzzle with the exact same sequence, where I could get exactly 4 copies of the sequence to spark a beautiful structure as is and make a full symmetric switch.

http://www.eternagame.org/game/puzzle/8658067/
Something I found much harder to make with the puzzle I made with 4 “1⁄2” sequences. I couldn’t make it fully symmetrically either in the second state.

The only difference between these two puzzles are that in the first I had inverted the sequence - to make it identical to the real life mRNA. So this is just another example of direction matters.

It is not as if nature designed these sequences with the thought that I would come along and just for the fun of it make switches of them. So I wonder about the why. Now I can’t help wonder if this is generally the case for the designs that seems to need to get an extra part of the sequence along to make pretty structures, that they are of the uninverted kind?

[Eli_Fisker]

RNA binding proteins binds to repeat RNA base sequences

Rhiju shared a super interesting paper. He described it as the following: an interesting paper on which RNA sequences are most likely to be bound by proteins in human cells.

Sequence, Structure, and Context Preferences of Human RNA Binding Proteins

He shared this image along. 

Here is the main idea of the paper: 

“We find that many RBPs (RNA binding proteins) bind a relatively small, defined subset of primary RNA sequence space that is rich in low-complexity motifs composed primarily of just one or two base types.” 

Rhiju: When I saw these sequences, I thought of some of your points about recurring sequence motifs (like magnet regions) in switches. Maybe biological proteins have evolved to toggle natural riboswitches inside our cells?

Eli: So you are thinking bigger proteins, not just smaller molecules or peptides. Perhaps this may explain why there is also a high ratio of repeat bases outside of the aptamer region in natural occurring riboswitches. In the “static” region.

The paper said they used several fixed aptamer sequences and mutated the sequence around them. I can’t seem to find out which aptamers they used. Anyway, from what I have seen so far, pretty much all aptamers carries base repeats and magnet segments. And so do their surrounding sequence.

Ok, so basically you are guessing that perhaps proteins can also interact with aptamers in our body, not just the molecule the aptamer is supposed to bind, right?

Rhiju: yes! actually these little sequences appear to act as aptamers for the *proteins* not small molecules.

Eli: Ah.
But it is the same principle. Just put in place two different places.
Lol, protein aptamers.
So RNA aptamers have repeat bases and sequence bias. Despite they are to bind with a wealth of different kind of molecules. RNA binding proteins seems to favor binding to RNA with repeat sequence and RNA with sequence bias.

rhiju: yes, it is indeed the same principle!

eli: I wonder what physical reason there can lay behind this?

rhiju: yes i wonder too
this is a frontier area in RNA biology
and in RNA medicine

eli: 
I mean I suspect that repeat bases let easier go.

rhiju: 
essentially every RNA in our bodies is thought to be bound (maybe even ‘coated’) by proteins
but we have few computational tools for modeling these RNP complexes
and even fewer ones for designing them

eli: 
Because I observed that for static designs, that if we used several GC pairs in a hairpin stem and if we didn’t cross (flip) at least some of them, then they would easier slide or mispair somewhere else.

(Tutorial: Crossed GC pairs in stems #1 - High crossing frequency)

rhiju: but that understanding will be crucial for designing therapies that, e.g., silence RNA genes

eli: The variation seemed critical to stability
Whereas well mixed bases seem detrimental to switches.

rhiju: exactly

eli: So I think something special happens when there are repeat bases. Something that will allow for easy bind but not too strong a bind. I recall I mentioned base size. That pyrimidines were short and purines being long. And having long bases side by side, and short bases side by side, made for an easier split. 

Background: Why do switch RNA like repeat sequence?

Idea for comparing eterna switch sequences with protein binding RNA sequences

Rhiju suggested that we could take a look at which of the switching sequences in the best eterna switches that match the list in the paper with the protein RNA binding sequences. 

[rhiju]

Thanks, Eli! What would help would be a bioinformatic analysis of some switch solutions (even in silico switches) to see which sequences are over-represented in the switch regions. Perhaps another player with some coding skills could help out?

[rhiju]

Also, one more comment – Nando has written the best bot at solving singe-state puzzles, called NEMO (https://www.biorxiv.org/content/biorxiv/early/2018/06/14/345587.full.pdf).  It requires having some approximate rules for what nucleotides to put in loops & stems.

Perhaps we could help Nando to make a NEMO for switches if we could collectively compile nucleotides (or stretches of nucleotides?) to guess for switch regions.

[Eli_Fisker]

In silico switches versus lab switches

Rhiju, you mentioned in silico switches also. 

There is one additional thing we need to take into account when it comes to in silico switches versus lab switches. 

Just as single state in silico puzzles are not the same as single state lab designs. In silico switches and lab switches are not entirely the same animal. Things that are legal for puzzles, will be illegal in lab. 

What will be an optimal base frequency for the switching areas in a switch will strongly depend on what shape the puzzle has. 

Asymmetric switches versus symmetric switches

An asymmetric switch - like most of Malcolm’s switch puzzles - will need an entirely different base frequency compared to symmetric switches - like many of Cynwulfs. 

• The symmetric switches demands heavy base repeat - be they in silico or in lab. 
• The asymmetric switches like Malcolm’s demand a solving style that looks more like a static design - as in more mixed bases. So AU base pairs like AUAUAU. (This is still a very repetitive sequence pattern though - it is the sequence that is repetitive and less the individual bases.) And some of their GC pairs in stems crossed. Which is also typical of static RNA designs.

I noticed this discrepancy between switch lab and switch puzzles, since I tried solve Malcolm’s puzzles in the style I had come to expect to work for lab switches with repeat bases and his switches puzzles utterly resisted my attempt. 

So what will be optimal base solving frequency for switches will heavily depend on the structure of that switch. 

I think we will find that just as there are static RNA shapes that are impossible to make stable in lab (the symmetric ones - if the puzzle is not very small) despite them being possible in silico. 

I suspect we are also going to to find that there are switch shapes that are impossible to make make switch in lab. (potentially the more asymmetric ones)

Position of switch elements in lab switches versus switch puzzles

One more thing that is done in switch puzzles, that may not be possible in lab: 

A lot of the puzzle switches that eg hold the FMN aptamer has the FMN aptamer very far away from the rest of the puzzle that it is going to interact with. In lab switches the switch elements seems to favor particular positions and a certain distance range in relation to each other. Also depending on if it is an ON switch or an OFF switch. 

[Eli_Fisker]

Hmm, perhaps this is the explanation on something I have been wondering about.

I had been wondering about the strong repeat base frequency in certain past static RNA designs that structurally were heavy in symmetry. 

So far I had explained this oddity in static designs with that long stems are more way more tolerant to any kind of pattern, repeat bases too, than shorter stems. 

I still think this is true. 

However there may be an extra reason. That heavily symmetric puzzles that are supposed to be static, may somehow benefit from repeat bases. Just as heavily asymmetric switch puzzles seems to benefit from well mixed bases. 

Judging from the amount of repeat bases and repeat CU and GA in particular this sequence ought to be a switch. 

https://eternagame.org/game/browse/24153/?filter1=Id&filter1_arg2=25205&filter1_arg1=25205

When it is not it may be because of how these repeats are used. They are made so the sections individually are making it only possible for neighbouring strands to pair with the intended neighbour and not the neighbour at the other side. 

This design was the top scoring design in The Cross - pre cloud lab. 

The design was still a winner in the cloud lab rerun. But other sequences were better solves. Some of the base repeat sequence disappeared in the new winners. 

Sum up of what a RNA design really wants

Since switch switch designs seems follow the opposite rules as those for static designs, so I could sum it up like this:

• symmetric switches want repeat bases and biased sequence - also in stems
• asymmetric switches want more mixed bases with some repeat sequence
• asymmetric static designs want well mixed bases in stems and some repeat bases - mainly in loops. 
• symmetric static designs want well mixed bases with a lean towards more repeat bases - also in stems
  While I think symmetric switch RNA and asymmetric static RNA are the ideal. 

[Eli_Fisker]

Downboosting around stems in switch multiloops 

One of Cynwulfs new puzzles caught my attention. 

This puzzle had only two areas left to solve. The static stems outside of the switching area and a small part inside of the switching area. It was the small part inside the switching area that specifically caught my attention. 

I have highlighted (black rings) the downboosting areas - bases that raise energy and make it more positive next to the stem they are placed around. 

The more switching an area is involved in, the more downboosting seems to happen in the switching multiloops. 

State 2

State 1

For lab switches there has long been a trend that the bases just around a switching stem - either at the bottom of the stem or for the two boost spaces in the loop, trends toward being downboosted. Having the bases that is added causing a more positive energy. 

Positive switch multiloops/Negative static multiloops?

As I have earlier observed the RNA folding rules for switch RNA are the opposite to those for static RNA. 

Basically if it works for static designs, it is worth disrupting in switch designs. If it works for switches, it is worth trying to disrupt for static designs. I have seen it again and again. 

For static designs, boosts that made energy more positive around stems in multiloops that are supposed to be static were detrimental to good stable RNA designs. This is one of the areas where the RNA algorithms failed big, because they didn’t seem to know. Vienna1 in particular.

I have noticed that for labs that making energy more positive for a multiloop (destabilizing) seemed to help in switch labs. However I haven’t seen downboosting for multiloops go on to the same degree yet as in Cynwulfs puzzle. 

With energy turned on, extremely positive loop and multiloop energy: 

Static designs favors higher negative energy inside multiloops - although it can get too negative also. Switch designs tends to favor higher positive energy inside multiloops in the switching areas. Although it can probably also get positive too. 

Perspective

I wish to see what Cynwulf did with downboosting multiloops in a switch puzzle, tried in lab. I will be very interested in if the puzzle pattern with multiloop downboosting can be transferred to lab.  

Background posts: 

I have given an introduction to downboosting here: 
Quick overview of the incoming results
Downboosting hairpin loops etc 

Multiloops - Static designs
Blue, green and red nucleotides in multiloop ring
Clean halo in multiloop ring or flaws of Vienna
Something loopy – energy in multiloops

Multiloops - Switch designs
Making a switch happen by weakening multiloop in unbound shape (in the switching area)