How Eterna players can generate 100,000 promising OpenCRISPR designs in just eight weeks

"Creating (puzzle) games within a (active lab) game?"

This bring up something I have wanted to do for a while but is harder with the current interface. 

I would love to make tutorial puzzles based on active lab designs like our current labs. I have earlier done so for labs that don’t have outside inputs. 

The puzzle maker is not set up to deal with labs with inputs, such as the current lab. This will probably take a lot of work so not realistic now. 

What I am really wishing for would be a script that could grasp the overall puzzle structure, so I could simply just cut out the bit with the DNA binding which I’m not interested in when it comes to tutorial making. 

I know there is a script that can grasp structures, but I have tested it this round and it is not working for me with these more complex puzzles. I have earlier used it to grab the structure of a lab puzzle. 

http://www.eternagame.org/web/script/2831764/

I know that I can recreate the puzzle from scratch, but I know I will get a lot fewer tutorials made this way.

Hi Gerry Smith!

I found your observations interesting. I think you have some real good insights here.

“Therefore the tendency to want to win at one level was a bias against cross level learning.”

You are absolutely correct. To learn what will potentially work in future labs, we have to be willing to take a loss on score. It is helping in the EteRNA world that we haven’t been getting lab coins for our designs for a long time.

The real game is figuring out how RNA folds. What it prefers.

If we can predict what it wants, we can help scientists make better tools for solving real world problems. 

You shifted your focus by will, to learn to overall improve your skills.

You did something else. You choose to improve your skills compared to yourself, which is really the best measure stick there is.

If we shift our focus to that there are people out there who are ill and that we actually hold it in our power to do something about it, I think it will be easier to drop the less healthy competition part. Which isn’t good for us either.

We are not winning this game for ourselves. We are really trying to win this game for humanity.

1 Like

This is an interesting question.  To the best of my knowledge, no one has actually done an analysis of lab results to see what predictive value there is in knowing which engines approve of a design. It’s certainly not for lack of data; we have plenty of data for the FMN/MS2 input/output combination. It would definitely be a very valuable thing to know.

I have too many ongoing commitments to take an active role in a project like that right now, but I could certainly help guide you to the relevant data/resources if you (or anyone else) wanted to pursue that.

Eli, what about the puzzle maker keeps it from making puzzles similar to the current labs.  It does have a facility for binding an aptamer to a loop.  It doesn’t have the ability to specify the separate DNA strand, but you could get essentially the same result by pretending it was a continuation of RNA with 3 or 4 A’s tieing the two pieces together.  (That’s actually what Nando’s hand-crafted lab puzzles do.)

Also, I just tried the script with (puzzle) id 8047255 and (submission) soln 8175708 and it ran.  The output of the script is quite verbose, but it looked reasonable and included a folding (which is undoubtedly from an earlier version of the Vienna folding engine.)  Does it fail in some other way for you, or do you just find that the folding doesn’t match any that the current engines give?

Omei, thx for looking into this. 

Here is a specific design I would like to turn into a tutorial puzzle and the ways I tried.

  1. Beam to puzzlemaker. (Right click in any puzzle or lab design and choose Beam to puzzlemaker) 

This is normally the easiest way to grab the structure for a puzzle, using the function Nando got built in.  

Result this time: 

  1. Use jandersonlee’s script to grab the structure. 

Puzzle id: 8047736
Solution id: 8143991

Input in script: 

Result this time: 

Library Loader use… true
title: “CRISPR/Cas9 - FMN 1a - Exclusion”
created: “07 Aug 2017”
object: “[{“type”:“oligo”,“oligo_sequence”:“GCGUCUCAUCUUUAUGCGUC”,“oligo_concentration”:1e-16,“oligo_name”:“Target DNA”,“oligo_label”:“T”,“secstruct”:”(((((((((((((((((((((((((((.((((…))))…)))))))…((…))…((((…))))(((((((…)))))))…&))))))))))))))))))))",“structure_constrained_bases”:[21,31,117,129],“anti_structure_constrained_bases”:[32,43,106,116],“anti_secstruct”:"…(((…(((…)))…)))…((…))…((((…))))(((((((…)))))))…&…"},{“type”:“aptamer+oligo”,“fold_version”:2,“site”:[34,35,36,37,38,39,40,41,108,109,110,111,112,113,114],“concentration”:2370,“oligo_sequence”:“GCGUCUCAUCUUUAUGCGUC”,“oligo_concentration”:1e-16,“oligo_name”:“Target DNA”,“oligo_label”:“T”,“secstruct”:"(((((((((((((((((((((((((((.(((((((…(((…)))…)))))))…)))))))…((…))…((((…))))(((((((…)))))))…&))))))))))))))))))))",“structure_constrained_bases”:[21,43,106,129]}]"
annotations: null

Plus a lot more.

I ran the script in a normal switch puzzle and got figured that the first two dot bracket structures were state 1 and 2, in that order. 

I notice there are some “&” in the dot bracket structure. Not sure how the puzzlemaker will deal with those.

I suspect the locked parts of the puzzle are seperated out and not included in the first sequence structures.

  1. When I add in these structures in the puzzle maker, I get this, which doesn’t look like the structure of my original puzzle. Also not if I add in my original sequence.

How do I get the structure from this to get to making a working tutorial puzzle? 
 

I can basically lose everything after base 130. 

Ok, I can actually get my structure from lower in the script.

[{“type”:“oligo”,“oligo_sequence”:“GCGUCUCAUCUUUAUGCGUC”,“oligo_concentration”:1e-16,“oligo_name”:“Target DNA”,“oligo_label”:“T”,“secstruct”:"(((((((((((((((((((((((((((.((((…))))…)))))))…((…))…((((…))))(((((((…)))))))…&))))))))))))))))))))",“structure_constrained_bases”:[21,31,117,129],“anti_structure_constrained_bases”:[32,43,106,116],“anti_secstruct”:"…(((…(((…)))…)))…((…))…((((…))))(((((((…)))))))…&…"},{“type”:“aptamer+oligo”,“fold_version”:2,“site”:[34,35,36,37,38,39,40,41,108,109,110,111,112,113,114],“concentration”:2370,“oligo_sequence”:“GCGUCUCAUCUUUAUGCGUC”,“oligo_concentration”:1e-16,“oligo_name”:“Target DNA”,“oligo_label”:“T”,“secstruct”:"(((((((((((((((((((((((((((.(((((((…(((…)))…)))))))…)))))))…((…))…((((…))))(((((((…)))))))…&))))))))))))))))))))",“structure_constrained_bases”:[21,43,106,129]}]"

State 1: (((((((.((((…))))…)))))))
State 2: (((((((.(((((((…(((…)))…)))))))…)))))))

(after I delete the first 20 bases and the bases after 130)

Here is the sequence I cut out from my original puzzle:
GUUUGAGAGCUACAGAGGAUAUACAUGAGCAUCAGCCAUGUAGAAGGCUGACAAUAAGUCAGAGGAUAUACAUGAGGAUCACCCAUGUAGAAGGCUGUAGCAAGUUCAAAUAAGGCUAGUCCGUUAUCAACUUGAAAAAGUGGCACCGAGUCGGUGCUUUUU

Using the Copy/Paste/Replace booster (Introduction here)

However due to my puzzle being a monster with two MS2’s and two aptamers, I have trouble stabilizing it, even after adding in an aptamer molecule.

How my puzzle looks in the puzzlemaker after the operation. Dead unstable.

While it has the structure of the original puzzle.

Do the static stem in the switching area have a function?

Back in the early Riboswitch labs I made an observation that a static stem would turn up in a huge amount of the winners. I started to speculate about that the static stem was necessary. That it had a function on its own.

For background see the section No Static Stem.

Particularly in the Exclusion NG 2 lab did this stand out. In our lastest riboswitch lab (101), roughly 95% of all the winners have the static stem right next to the aptamer gate.


http://www.eternagame.org/game/browse/6369184/?filter1_arg1=6456051&filter1=Id&filter1_arg2=…

I ended up doing some experiments, where the space for the static stem got deleted. All designs I did this to across labs, ended up with a lower score than the original designs except in one case.

Here are the two small loop labs that are equivalent to the Exclusion NG 2 lab. All these designs are based on winning Exclusion NG 2 designs, but without space for and the sequence of the static stem. They all took a score hit.

**

**


 

Omei’s insight + mine = Is the Static Stem in the switching area doing Coaxial stacking?

What I’m really interested in now is getting something specific tested.

Omei has earlier brougth up that stems when being next to a sequence that would bind up with an outside input, were doing coaxial stacking and helping the input bind with the sequence next to the
static stem.

Intro to flush stacking energy

I have started to wonder if not the static stem in the switching area of a riboswitch, when next to the aptamer or MS2 is doing coaxial stacking.

Combining these two ideas

  1. Static stems are (often) necessary in the switching area
  2. A coaxial stacking aids the binding of the sequence next to it

And it would pretty much make sense that static stems do have a function.

Static stems seem necessary at specific spots and those specific spots are somewhere where coaxial stacking will help the fold.

Potentially allowing for a coaxial stacking of the static stem with the aptamer gate and as such potentially aiding the bind of the FMN molecule.

The Nature of Static Stems

I basically think static stems are a bit like mushrooms. If there are enough bare soil, more will pop up. So if there are excess bases not in use or needed for the switching area itself, a static stem will come in handy.

If there aren’t excess bases, a static stem isn’t needed. If there are lots of excess bases, more static stems are needed.

I also think that if there is space for two shorter static stems, that they will be better than one real long static. Basically RNA seems to not be too happy about all too long stems. Unless some GU or mismatches are stuck into them. Also if coaxial stacking is occouring, then two static stems would mean two energy bonuses. :slight_smile:

Additional experiment proposals

While I have gotten closer to have my experimental starter designs put up, I realized that in some labs, the aptamer took so much place or the switch elements were placed in such a way, that I can only run 1 out of 4 experiments. Thus I have 90 slots to spend for other experiments.

What I’m thinking of is to use them on a design with 1 static stem. And do something like change the position of the static stem, so it in 30 designs will be next to the aptamer gate, in 30 be next to the MS2 and in the last 30 be inbetween.

It could also be interesting having frequency experiments where the static stem gradually gets moved between being next to the aptamer gate to being next to the MS2. In a systematic manner. Like the ones Omei has done and I show in this intro to using eterna spreadsheets.

In particular in the CRISPR/FMN NG labs, this should be interesting to test as we can literally stick in past winners. And then move around the static stem.

Get creative… :slight_smile:

1 Like

@eli I’m not that conversant with the puzzle maker.  But if I enter the following two structures:
(((((((((((((((((((((((((((.((((…))))…)))))))…((…))…((((…))))(((((((…)))))))…))))))))))))))))))))
and
(((((((((((((((((((((((((((.(((((((…(((…)))…)))))))…)))))))…((…))…((((…))))(((((((…)))))))…)))))))))))))))))))
and assign the bonus to the appropriate loop, I see

which seems like the right structure for you.  If you fill in the proper bases and it doesn’t behave the way you expect, things may just not work the way I think they do.

1 Like

Despite increase in hangups, I want to give huge kudos to developers of mutation tool.  Without a doubt it has vastly increased my value as a lab participant in being able to quickly explore many unknown areas and finding designs with multiple engine hits.  Makes it so much more fun to be able to look for interesting patterns and explore further. 

1 Like

Thx Omei, for thinking about this workaround.

Unfortunately puzzle maker do not agree. I got the sequence and guide DNA sequence added. My puzzle is still unstable.

“Unstable”, as commonly used by players, simply means that the folding engine’s calculation of MFE for the sequence doesn’t match the target structure.  I may be missing your point, but can you resolve your problem by simply adjusting the target structure to match the predicted folding?  It looks like you took my target structures as gospel, instead of just being a model for the locked bases.

Ultimate Symmetry Continued

My Ultimate Symmetry series in Exclusion labs holds double aptamers.

I put in rotated aptamers with the intention of getting the aptamers turning their favorite orientation towards the switching area.

However I have realised there may be a tiny problem… :slight_smile:

Namely that if the aptamers are placed like this and the rest of the puzzle scaffold are not strong enough to hold the aptamers, the following may happen.

Rotated aptamers

https://drive.google.com/open?id=0B55atkAMoafyVEE0eUxpM1BTWXc

  • The aptamers may stay ON in both states.

  • The aptamers may instead of shutting off, find each other’s matching FMN half, do an aptamer rescue mission, so both the aptamers will stay turned ON.

I have earlier called the kind of aptamers for mirror aptamers. But really the extra aptamer I added is rotated 180 degrees in relation to the locked aptamer.

Sameturning aptamers

So I have started to suspect that we will have a better shot off getting the Ultimate Symmetry designs to work, if we instead make Same turning aptamers.

https://drive.google.com/open?id=0B55atkAMoafyX2REYzljeTg2NGs

One more Experiment

I think this call for another experiment pitting two such sets against each other. :slight_smile:

#sameturningaptamer versus #rotatedaptamer

Sameturning Ultimate Symmetry versus Ultimate Symmetry

Biased Guide DNA

I wonder if the guide DNA chosen for the CRISPR labs have to do with anything specific? I wasn’t able to look up the sequence. Tried it backwards too. :wink:

I couldn’t help notice that the guide DNA for the CRISPR labs have a strong sequence bias. The target DNA has CU (pyrimidine) bias, whereas the RNA part of it has GA (purine) bias.

I happen to like it. It is very switch like. I won’t be surprised if there is some bias to where CRISPR likes to land.

Why do switch RNA like repeat sequence?

Back when switches were new in lab, they caused me wonder as to why they would carry so much repeat sequence. Turned out the repeat bases in the design came from the locked bases in the FMN sequence. The aptamer itself carried repeat bases and as such sparked repeats.

However I found out this was not an isolated incident just relevant to the FMN aptamer.

The sequence repeat seems central to switch elements like aptamers, MS2’s, microRNA’s and switch inputs in general.

Long and short RNA bases

I think there is a specific reason for this. I think it helps them switch… :slight_smile: Bear with me.

There are long bases and there are short bases. The short bases are Cytosine and Uracil. The long bases are Guanine and Adenine.

Why do basepairs pair up?

RNA behaviour explained with Lego

I have found a new and better way to illustrate my point that RNA switches are having certain repeat bases and why.

Static RNA likes its base pairs rather well mixed up. Switch RNA is different.

Switch RNA tends to hold rather specific repeat sequences. 

https://drive.google.com/open?id=0B55atkAMoafySEdlRjg4d1ZNNmM

At least when it comes to lego, it is obvious why the mixed bases at the left bind much better together.

Whereas there are little except hydrogen bonds to hold the two RNA stacks at the right, together. I guess the bases can stack better when they are of similar size. So for the switch bases, they can bind through hydrogen bonding, but them not truly mixing,

Haven’t had enough Lego? Read Ksteppe’s beautiful introduction to energy in the RNA world.


Understanding Free Energy (using Legos)

Background posts

For the background story on repeat sequence in switch elements see:

Adventure in Riboswitch Wonderland

Here is the backstory of finding the periodic repeats in the early EteRNA Switch labs:

Can periodic repeats in RNA switches be programmed?

Gerry Smith sent me a question about a design. He has allowed me to share our conversation for the benefit of all.

Hi Eli -

If you have time, take a look at these two designs. Zama sent me the first. I liked it because it had minimal movement.

So I added another static stem closely on the other side of the aptamer. So the aptamer is closely bordered by two static stems.

Also the folds & movement in all three engines are relatively the same.

I am trying to employ what I understand from your observations to improve designs I come across.

Is the altered design an improvement?

thanks Gerry

Zama design
http://www.eternagame.org/game/browse/8047753/?filter1_arg1=8220506&filter1=Id&filter1_arg2=8220506

Altered design
http://www.eternagame.org/game/browse/8047753/

My answer:

Hi Gerry! 

Yes, you did improve the design by tying some of all the excess bases away. However your design and Zama’s are likely not going to have the MS2 turned off. 

In this case minimal movement is likely not going to be good. (But keep looking for it other places) 

This is an exclusion design. What they usually need (Talking of FMN that I know) is the following

  1. MS2 next to one of the aptamer sequences. Check - you got that!

MS2 is strong, so just the opening base pair splitting will not be enough to turn it off. The MS2 molecule binds at the hairpin loop. So it needs a helper sequence to get shut off. 

  1. A turnoff sequence that is right after MS2 and preferably targets an identical sequence stretch that is present in both MS2 and in the aptamer. This one both of you are missing. 

Best way to show this is to check this past winner: http://www.eternagame.org/game/browse/6369184/?filter1=Id&filter1_arg2=6456051&filter1_arg1=6456051

Swap between the states to get a feel for it. In particular notice the base stretch at 44-48. This is the turnoff sequence. 

Also take a look at this puzzle again: http://www.eternagame.org/web/puzzle/8057283/

Thanks for tool (booster) it gives solutions fast , but in general this only makes same structure foldings. So we can find best solution of some structure , but in other way of making best score solutions is to make many of structures instead (I believe it gives better result than thousands solutions of same folding structure) As i see today (after i started to use booster) there is coming a problem that 150 solutions for player is not enough.

And I have a question about OpenTB … What our gamers give to medicine ? Is any company started to make a drugs that can heal tuberculosis or it is just for make gamers to believe that they do an important work and it doesn’t give any result ?  

I agree with worseize so I don’t use it. I get one to work, mod it 4 or 5 times to obtain different structures, then reset and go at it again. I’m maxed out so you might think of increasing the maximum allowed.

@JR I admire your approach to labs – trying different ideas, with only a few variations, and still managing to fill all your slots.  Good work!

As for increasing players’ quota of slots, it’s definitely on the table for consideration.  But there was a rationale for setting it where it is.  Even at 150 slots per puzzle per player, it would be possible for about 20 proficient players to fill up all the experimental slots.  A few, like you, have the dedication and patience to generate that many designs without any script assistance.   But there are even more who could simply use their familiarity with scripting to generate however many designs they wanted.

Basically, we wanted to try to level the playing field between the few technically proficient and the rest of the players.  It is a two-pronged approach.  First of all, to give all players some access to automated tools, but at the same time try to diversify the submissions by encouraging as many different minds as possible to tackle the problem from their unique perspective. Hence the per/player quotas were set relatively low.

We’ve never been very good at predicting in advance how many players and how much effort will go into a specific lab round, and so there is always some guessing at the beginning.  Some times quotas get lowered as time goes on, and sometimes they get raised.  Sometimes deadlines are extended and sometimes unused synthesis slots are transferred to a different experiment.  In all probability, one or more of those will happen in this round as we get closer to the deadline.

So thank you for your input, and most of all, your dedication!