What did you learn from: Lab 103 Round 1?

  • 6
  • Question
  • Updated 9 years ago
  • Answered
When we enter the lab, we are expected to use our knowledge of past labs to inform our votes. So... what did you learn from this round of synthesis?

You certainly don't need to comment on each synthesized design. But if you learned something specific from one of the synthesized designs this time, share it with us! What did you learn?


  • Grom 103 v2, Grom, 77/100


  • Bulgy 103, singinst, 82/100


  • JestersDaughter1, dimension9, 83/100


  • bulged cross-1, Vexelius, 86/100


  • OccamRazor, Chesterfield, 86/100


  • Very Best of Three Designs, madde, 87/100


  • Mat Bulge Cross (relook at 102), mat747, 89/100


  • -45.5 kCal 107 M.P., donald, 94/100

Photo of Chris Cunningham [ccccc]

Chris Cunningham [ccccc]

  • 97 Posts
  • 13 Reply Likes
  • excited

Posted 9 years ago

  • 6
Photo of Chris Cunningham [ccccc]

Chris Cunningham [ccccc]

  • 97 Posts
  • 13 Reply Likes
Here's a question to start off the discussion: why did this happen in Mat Bulge Cross (relook at 102), mat747, 89/100??
Photo of AnticNoise

AnticNoise

  • 11 Posts
  • 3 Reply Likes
Even more interesting is in JestersDaughter1, dimension9, 83/100. The first 5 pairs from right to left are identical and set up in an identical fashion but the bonding was successful.

Maybe this implies that the high difference between those two stacks of 1.8 kCal decreased its chance of a successful bond?

Theory posted below...
Photo of dimension9

dimension9

  • 186 Posts
  • 45 Reply Likes
Yes, Chris, and I have also been baffled because in my previous design, not one, but TWO G-C pairs came apart! WHAT? Of all places to break, how could it break on the two G-C pairs !!!??? I have been questioning the Devs about this, but have not really found anyone yet who had a definitive answer.
Photo of dimension9

dimension9

  • 186 Posts
  • 45 Reply Likes
Chris, Most AWESOME Post! - now if people actually go to the lab and watch for themselves and analyze how and where and why all these broke when switching to natural mode - it will be super-instructive! And we will begin to see better and better designs from everyone!

Huge Kudos to you for taking the time and effort to make this incredible post!

-d9
Photo of dimension9

dimension9

  • 186 Posts
  • 45 Reply Likes
What I learned: (from the notes to my JestersDaughter2 entry)

Jester's Daughter scored 83 in round one - not the highest score of the round but, it WAS one of the less-deformed of all the synthesized entries when switched to natural mode.

After analyzing what happened to mine, I think the problems were:
1) The G @ 50 was more attracted to the C @43 than to it's paired base (U) @ 41 (because G's are more attracted to C's than U's)

(Note: I had not even thought to watch out for this during the first construction of the design -THIS is the BIG LESSON I took away from this round - we MUST be alert for alternate bonding possibilities, and try to eliminate them from the START) ...and additionally, we need to begin to learn how to recognize the ways in which a design is likely to deform itself in order to reach a lower energy configuration - How far will it go? How far CAN it go?)

2) Because the U @ 49 also had an alternate matching base A @ 44 besides its paired base A @ 42, it too could move

These two double alternate bonding opporunities enabled a configuration with a lower overall energy to form - so it did!

To try to fix it, I flipped both the A-U @ 42-49 and the U-G @ 41-50 removing (hopefully) both these alternate bonding possibilities To help keep this in place, I mutated the G @ 41 to an A to make that bond stronger.

In Madde's very similar design, the U @ 51 moved over 5 whole bases to bond to the A @ 46 creating 2 smaller lower energy loops - (and shoving everything in between right out of the way to do it!).

In my design, to try to prevent this, I subsitiuted a G @ 46 (to replace that A). This should act to create a weaker attraction there, and prevent that U @ 51 from moving over to bond to it as it did in Madde's design.

-d9
Photo of Chris Cunningham [ccccc]

Chris Cunningham [ccccc]

  • 97 Posts
  • 13 Reply Likes
Here's the image for his #1, underlined, bold point:


I think conciseness and prettiness are going to be more important than content for these posts, if we expect anyone besides us to look at them.
Photo of dimension9

dimension9

  • 186 Posts
  • 45 Reply Likes
Chris, thanks for adding the graphic! I need to begin to do that; also, point taken on being more concise - I believe you are absolutely right :) -d9
Photo of cesium62

cesium62

  • 8 Posts
  • 2 Reply Likes
What I would want to try next is to see if some of the large sub-pieces remain stable when attached to other large sub-pieces. The right side of the cross is the same in madde's and dimension9's. Splice that with donald's left side of the cross. I haven't looked at the differences of the top and bottom bars.

Apologies for gross speculation, but...

For madde's vs mat's... mat has a GGCAUCUUG sequence on top of a CUGUAGAGC sequence. The AU and UG in the top row might tend to pull together and spread out the pieces in between? Similarily for the UA and GC in the bottom row. Then the U and G that we wanted to bind together don't bind well because the stacks have been stretched thin at that location?

Madde had a GGUCUCUUG sequence on top which has a lighter GU pair further down the stack. On the bottom, the CCAGAGAGC sequence just has the one GC pair on the bottom. So the top stack would be lightly pulled apart, the bottom stack hardly pulled at all, and so the desired GU bond gets to form??
Photo of cesium62

cesium62

  • 8 Posts
  • 2 Reply Likes
Here's a stab at a pictorial representation.

G GC AU C U UG
C UGUA G A GC

G GU C U C U UG
C C A G A G A GC
Photo of Chris Cunningham [ccccc]

Chris Cunningham [ccccc]

  • 97 Posts
  • 13 Reply Likes
Anyone with some Chemistry knowledge able to give us a clue on whether this has merit? My first guess is that a G and a C next to each other on the same strand don't attract each other, but my chemistry is in the past.
Photo of Chris Cunningham [ccccc]

Chris Cunningham [ccccc]

  • 97 Posts
  • 13 Reply Likes
I'm sorry that no one with more chemistry knowledge stopped by to answer, so I'll give my understanding of it.

The fact is that Us, As, Cs, and Gs all use the same kind of "attraction" (called Hydrogen Bonds) to get close to each other, and the reason A-attracts-U is because the specific hydrogen-bonding locations on the A line up well with the hydrogen-bonding locations on the U when they are oriented so that they are in an RNA strand. There is nothing instrinsically attractive between an A and a U any more than between an A and a C.

So, my limited chemistry knowledge tells me that C's and G's that are adjacent in the same line should not really interact at all -- their hydrogen-bonding sites are on the ends, and they probably don't pull toward each other. However, I was really hoping someone more qualified would answer this question.
Photo of AnticNoise

AnticNoise

  • 11 Posts
  • 3 Reply Likes
Mkay so this may take a while but I'll try and do it in a comprehensible way. So last night I started looking at these and analysing any discernible differences between the places that broke and such and found what may be a plausible and very interesting explanation for the mis-folding that occured. Dimension 9 feels the same way and we discussed this last night. This might result in an updated codon reference sheet based on energy difference and distribution across stacks.

The point that I noticed between all the best designs where only 1 or a few bonds didn't properly align and form is that they all have high energy gradients between the stacks that support them. I'll go through each one accordingly and you can all tell me what you think of the idea.

First of in donald's. The only part that didn't bond was the final G-C bond on the big loop. But if we look there is a significant energy gradient that we don't find in the other designs where that particular bonding was succesful:

As you can see there's a 1.2 kCal difference in energies between that stack and the loop. In all the other cases that difference was never more than 0.5 kCal's.

Next in mat's. The part that didn't bond was the U-G bond just after the large loop. If we look into detail we see that there is a large difference between the stack energy of the stack that didn't form and the stacks on either side:

However the first 5 base pairs from right to left are the same within dimension9's design where the bonding was succesful and in madde's the energy difference is also apparent but the bond also successfully formed. This either means that something about the overall structure is affecting this or that the energy difference means that that bond has a higher chance of not forming.

Madde's. Within madde's the bonds that didn't form are just after the bulge on the left side of the structure. This is very similar to mat's in that the stack differences on either side are very high:


Finally we have Vexx's where two areas bonded badly. If we take the bulge on the left first we see that the bonds following it didn't form properly. The interesting this is that the loop in this case has a positive energy value. There is also a another gradient occuring:

And in the second place in the central structure the G-C bonds mis-bonded and chose other sites to bond but if we analyse the details we see this:

In the other versions where this was successful there wasn't such a high difference between the energy of the stacks on all 4 sides. I think that the bonding may have changed because though that formation yields the best energy output another previously available bond option also yields a high output and sets it in cement without the chance to opt for the better energy option.

So those are my thoughts on something that may be influencing successful bonding within these structures... Thoughts, ideas or criticism? All are welcome.
Photo of dimension9

dimension9

  • 186 Posts
  • 45 Reply Likes
Antic,

Wow, NICELY done!, great, complete explanation! My only addition for the moment (following my own further explorations last night) is that I now suspect that not only codons with very large differences between the Quad Energies seem more likely to fail, but also those with very, very LOW differences between the Quad Energies. In chat last night, you had tentatively identified a difference level of 1.2 kcal - or over - as being as level where problems bonding may begin to occur. In my explorations last late last night after our chat session, I also seemed to notice that those with differences of 0.3 Kcal - and under - may also have this same issue. This raises the theory of there being a an optimal "window" with both upper and lower bounds to insure bonding success. - more on this later. Great post!

- d9
Photo of Chris Cunningham [ccccc]

Chris Cunningham [ccccc]

  • 97 Posts
  • 13 Reply Likes
I ran some numbers and I have to say I don't agree with these theories. I went through Vexx's setup and, ignoring bulges and such, made a list of all the negative-energy stacks in order. Then I made a list of all the differences in neighboring stack energies.

If a high/low difference-in-stack-energy caused problems, we would hope to see some clue of it here, but I think there is none:
0 OK
0 OK
0 OK
0.1 OK
0.2 OK
0.2 FAILED
0.3 OK
0.3 OK
0.3 OK
0.4 OK
0.5 OK
0.5 OK
0.7 OK
0.8 OK
1 FAILED
1 OK
1.2 FAILED
1.2 OK
1.3 OK
1.5 OK
1.5 OK
1.8 OK
1.9 OK

If you can come up with a coherent way to consider the energies of things like loops and bulges so that we would have a testable theory, I would gladly run some numbers on it, but honestly I don't understand the system you're using for dealing with those.

An alternate, simpler theory, is that complicated spots in a desired RNA shape often involve some positive-energy areas (loops, bulges) and some negative (stacks), and those have large energy differences.

I'm glad to keep considering the idea, because I'm sure I'm not seeing everything you guys are seeing.
Photo of AnticNoise

AnticNoise

  • 11 Posts
  • 3 Reply Likes
I'm wondering in that case if it's the presence of bulges and loops that is making this effect notable and of importance. One point though that I was making is when there are significant energy differences either side of a stack. For instance in both Madde's and Mat's.

However that being said I've been told that hundreds if not thousands of molecules based on each design are created within the lab and under that condition I'm stumped because Mat's set-up that didn't work is identical to both Madde's and Vexx's.
Although that being said Mat's only shares 5 base pairs with Madde's and Vexx's whereas Madde and Vexx share all 9 base pairs in that section.

I'm sure something of some significance is happening here unless the pattern is a complete fluke, I'll investigate a bit more but the round 2 synthesis results should prove interesting.
Photo of Chris Cunningham [ccccc]

Chris Cunningham [ccccc]

  • 97 Posts
  • 13 Reply Likes
Antic: I had an idea today: if you wanted to get some support for your theory, you could look at the round 2 submissions and predict which ones will fail in which places!
Photo of Adrien Treuille

Adrien Treuille, Alum

  • 243 Posts
  • 33 Reply Likes
ccccc: I think this is such a good idea!
Photo of Adrien Treuille

Adrien Treuille, Alum

  • 243 Posts
  • 33 Reply Likes
Hey guys. This is very interesting and got met thinking.. would it be useful to have a display like this in the game? More generally, what would be useful displays to have in the lab?
Photo of Chris Cunningham [ccccc]

Chris Cunningham [ccccc]

  • 97 Posts
  • 13 Reply Likes
If there was some reasonable in-game way to have a civilized discussion about each synthesized result (or even each submission!), that would be amazing. I have no idea how to implement such a thing without a significant moderator effort, but there is a chance your community would provide the effort. A place to look for inspiration would be the shirt.woot.com derbies: http://shirt.woot.com/Derby .

The first priority in this direction is to be able to identify "base #41" in the lab without counting by hand. This could be toggleable, and would make the images I created above much easier to discuss. Of course the bare minimum would be a mouseover of a base providing its number.

Also I would love to be able to link to a specific lab design (so I would have put 8 links in the original post above), but I don't know what URL would possibly point at a specific lab design.
Photo of Madde

Madde

  • 45 Posts
  • 17 Reply Likes
Just add a slider to decrease the bonding forces between intended pairs (whereas forces between unintentional pairs would remain constant) as I suggested in http://getsatisfaction.com/eternagame...

This would roughly simulate interferences and disturbances of the real RNA folding process we don't know or don't understand yet.

- lower %-value of bonding forces while the design is still in shape means more tolerant against disturbances

- you could add the %-value to the table of lab designs
Photo of Dave MacDonald

Dave MacDonald

  • 1 Post
  • 0 Reply Likes
If you could associate nucleotides to their number, AND to a list of comments... Also, it woudl be very useful to see the bonding energy in the synthesized results (the same energies reported by the game)
Photo of Chris Cunningham [ccccc]

Chris Cunningham [ccccc]

  • 97 Posts
  • 13 Reply Likes
Here's a Thing I Learned:

You've probably seen a long strand of identical bases "slide" before when you moved from target mode to natural mode, like this:


It happens when you have too many of the same base in a row: this allows for things to "slide" without upsetting any of the nearby structure. If you have a sequence of UAGUA and it slides, it will not match up with the things across from it anymore. A sequence of GGGGG can slide and still match.

In Occam's Razor, there was a sequence of two CGs in a row, and if you followed the Cs, there was a third C right next door.

The result was that it gave the opportunity for the edge with the three Cs to "slide" and still form most of the intended bonds.

What I've Learned is to avoid too many of the same base in a row, especially near complicated features -- near those features (bulges or pentagons or whatever), the strand of duplicates can slide and mess up my intended shape.
Photo of dimension9

dimension9

  • 186 Posts
  • 45 Reply Likes
In reference to Chris' comment above:

"The first priority in this direction is to be able to identify "base #41" in the lab without counting by hand. This could be toggleable, and would make the images I created above much easier to discuss. Of course the bare minimum would be a mouseover of a base providing its number."

This has already been requested for quite some time), and has been on the Developer's TODO list (#225) since 11/26/2010 - The Devs have been extremely busy since the opening, but I am certain they will get to this just as soon as they are able:

http://getsatisfaction.com/eternagame...

-d9
Photo of Chris Cunningham [ccccc]

Chris Cunningham [ccccc]

  • 97 Posts
  • 13 Reply Likes
You can put your comment below mine by going to Adrien's post, and hitting "Add Comment"
Photo of Daniel Cantu

Daniel Cantu, Alum

  • 6 Posts
  • 14 Reply Likes
Hey, everyone, you are doing a great job working on this. I think you have some great hypotheses going and it'll be really interesting to see how these strategies work in future lab puzzles.

As you can see, computers still have a difficult time predicting how RNA is going to fold in real life and we can use all the discussion we can get to help improve them.
Photo of cesium62

cesium62

  • 8 Posts
  • 2 Reply Likes
To what extent do we think we can cut and paste different segments that work to get an overall working design? Reviewing the lab results tonight, it appears that to get the right side of the cross to close, we probably want to use the arm that is used in "Very Best of 3" and a couple of other synthesized designs: GGUCUCUUG
CCAGAGAGC
with a short dangling strand of AAAGG in the top right.

Searching for "ggaaaguucucugg.*ccagagagc" I think that proven sequence didn't get selected for round 3. ('rifle' proposed a sequence with this subsequence which he stated was modified from the winner of the round 1 competition.)

Do we have a place where we should chat about why we are voting for whatever sequence? This chat was the closest I found after my limited search.

At the risk of immorally blatantly plugging my own sequence, I submitted "cut and paste" which splices the working left side of the current leading candidate onto "very best of 3". I then added a few decorative G's on the loops because the game teaches us we should do that and I didn't see obvious evidence that this hurt the folding. Anyway, I'd be interested in following a discussion of suggested sequences and why they may or may not work.

Of course, if one of the current round 3 selections gets a score better than 94, forget I was ever here.
Photo of Adrien Treuille

Adrien Treuille, Alum

  • 243 Posts
  • 33 Reply Likes
cesium62: I love that idea. I just voted for it, if only to encourage others, get it voted up, and see what the synthesis result is.
Photo of Samuel Johnson

Samuel Johnson

  • 4 Posts
  • 2 Reply Likes
I agree completely! Cutting and pasting sections that have folded correctly in the past is a great way to test why a section will work in one design but not in another.

Along those lines, I used only sections that folded correctly in round 4 to create an entry for round 5. I used the right leg from 1337 by Fabian, the top and left legs from Donald's nephew 2 by madde, the bottom leg from -45.5 kcal 107 M.P. by donald, and joined them with the center cross from Ding's mod... by Ding. None of the sections had to be modified to fit with each other so every part of this combo design has folded correctly in round 4 synthesis.

Even if this combo design fails to fold correctly it will be very informative to compare the places that worked before and not a second time.

Check out Ankh Will Fold! to see what you think. And yes, shameless self promotion, Please give it a vote! The synthesis results could be very enlightening.