Cool patter in 202 labs melt profile the winner look like a baseball cap.

  • 5
  • Article
  • Updated 9 years ago
I have been using nupack to look at the lab results trying to find some patterns so i can better design my labs. I have seen consistent patten of the melt profile that looks like a baseball cap.
Mat - Branches V2


Modulus


Twosies - 2 Jack and ball


Berex Branches 3-1


Berex Branches 4-1 is in the middle of the pack


and the last one Vienna bot design 7


Do you think this could be a predictive tool?
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes

Posted 9 years ago

  • 5
Photo of rhiju

rhiju, Researcher

  • 404 Posts
  • 123 Reply Likes
This is super-slick. I think this feature (which is sometimes called 'cooperativity') is also intimately related to some of the other patterns that players are discussing where individual elements should have similar stabilities ('energy smoothness', etc.) -- that also helps guarantee that the features 'melt' at the same temperature.

In an early version of EteRNA, we had these melting curves show up when you clicked the eyeball -- Jee can you put this back in?
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
this is the pattern of my design Terminator 81c I would like see if it is better or worse in synthesis
between 27c and 107c


and between 27c and 157c
Photo of alan.robot

alan.robot

  • 91 Posts
  • 36 Reply Likes
Very nice! In principle, the "sharpness" should be somewhat related to how negative the free energy is, see:

http://cdn.idtdna.com/Support/Technic...

But as Rhiju said, it probably contains more useful information since this is more complicated than just a single duplex as the DNA people (above) are used to thinking about. . .
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
I will have to read this pager a couple of more time but these paper is talking about a more complicated chemical reaction than a RNA whit 110 nucleotide folding on itself. two DNA strains with 50< nucleotide sticking to gather to form a target shape I am glad to see that i had observed something that the experts have seen. That they are using both of the melt profile and the change of in Gibbs Free Energy(in units of kcal/mole) to in their words to qualitatively assess the tendency to be potentially ‘bad’ for the formation of secondary structure or undesired duplex formations under given experimental conditions. but its getting late time to go out
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
Can the melt profile identify RNA shapes? Guess which one fail to form the target shape.



Berex star 3.1



Ding's imp'o'imp


1 - Donald's star
Photo of alan.robot

alan.robot

  • 91 Posts
  • 36 Reply Likes
I can attest to the fact that the real-world experiment that corresponds to these melting profiles is a very informative experiment (also really easy to do). I wanted to link the PDF to this article

http://onlinelibrary.wiley.com/doi/10...

but my current institution doesn't have access. Message me if you'd like a scan, I know one of the authors so hopefully I can snag a copy in the near future
Photo of alan.robot

alan.robot

  • 91 Posts
  • 36 Reply Likes
Here's the link:
https://docs.google.com/viewer?a=v&pi...

The executive summary is as follows: Using UV-vis spectroscopy, one can directly measure a melting curve analogous to what Josh has calculated using NUPACK. The standard way one analyzes this is to take the first derivative, which should reveal one or more distinct "peaks". If the RNA melts in a modular fashion (note that is an assumption) then each "module" will have it's own peak centered at the Tm of that section, and the overall first derivative of the melting curve will be the sum of all these individual peaks lumped together (this can be numerically extracted if the peaks are close together, if they are far apart they will be immediately obvious). This paper includes a nice example of yeast tRNA melting, which is deconvoluted into 3 separate transitions. tRNA has 4 arms, so it's tempting to assign each peak to roughly one arm, but in reality it may be a more complicated process than that.

http://en.wikipedia.org/wiki/Transfer...

Note that, of course, algorithms like NUPACK (and ViennaRNA/Eterna) can't predict tertiary interactions, whereas they are of course present in the experimental data. However, assuming each peak corresponds to an individual element of secondary structure, the predicted Tm's and free energies from prediction algorithms for each individual "module" should be reasonably accurate.
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
Wow Alan that is great stuff. It will take me weeks to fully comprehend all of this paper but they are taking the melt profile a lot deeper. I am just trying to id good folding sequence out a bunch of sequences with only 16 data points. Do you have any insights as to how they are using this in the real world placations?
Photo of alan.robot

alan.robot

  • 91 Posts
  • 36 Reply Likes
yes! This is used in alot of real world applications.

First, I should mention that the experimentally measured free energies (like the ones used in the creation of folding algorithms like the one in Eterna) are all extracted from thermal melting curves.

Second, many very useful molecular biology techniques like PCR and the many extensions of that technique rely on knowing the precise temperature you have to heat a sample to force everything to be single-stranded (unfolded) so that they can be read out and copied by enzymes like polymerase and reverse-transcriptase.

Lastly (off the top of my head, I'm sure I'm missing other important applications) an acquaintance of mine was using this technique to look at how RNA's from microbes that live at very high temperatures (thermophiles) are adapted to stay folded while they do their work. . . a cool example I thought.
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
I would love to read it.
Photo of Adrien Treuille

Adrien Treuille, Alum

  • 243 Posts
  • 33 Reply Likes
I love these plots. I wonder if there isn't a way that we could link to the NUPACK server directly from EteRNA, kind of how josh did it in his submitted designs.
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
Just having some fun. If anyone knows of any other algorithms that can model melting profile please send me their link.

CGUAUGAGAGUACA



CGUAUGAGAGUACAGUAUGAGAGUACA



ACGAUACAGUAUGAGAGUACAGUAUGAGAGUACAGUAUCAA

Photo of Eli Fisker

Eli Fisker

  • 2273 Posts
  • 515 Reply Likes
@Joshua, this post is very interesting.

You commented on one of my designs, Mat with robotneck and tetraloops, as you thought it looked like a wired hat like slope its a tweener - which is different from the others you took a look on.

You are absolutely right - it does look weird. I don't know if it's good or bad, though I lean towards the last.

Link my design:
http://eterna.cmu.edu/game.php?myType...

So I ran following designs through the Nupack analysis tool
(http://www.nupack.org/partition/new)

My design

Bot design 3, from which I borrowed the neck from,

Mat's original design V1, which I was doing a mod of with tetraloops

Mat's design added just tetraloops (to check if tetraloops did a difference from his original, thereby causing the shift in the melt profile curve)

From degree 27 to 47 the curve of my design resembles that of the human designs, after 47 the curve rises quickly as in the robot design from which I borrowed the neck – which also happens in the vienna example you brought further up in this post.

If I ad tetraloops to Mat's design to see if that does anything to the curve, it goes totally flat in the beginning. First I tought my design behaved like this, because I had borrowed a robotneck. Because the early rising pattern looked more like the robots.

Then I realised that I had placed one red nucleotide at position 4, and start wondering if that did a difference.

Here is my design without the 4'th nucleotide being red. Result: flat beginning, first rising at 57 degrees. Now it resembles the human designs more, even though it has a robotneck.



Here is how my design looks with the red nucleotide at position 4. A big change with just one nucleotide different.



My design is still in synthesis. But I am not sure adding that red nucleotide was a good idea. :)
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
I don't know Its a new theory. I have to make lots of observations prove that it has any merit. I am just using the system to id similar designs kind of like a targeting system . I have no Idea how to us this theory to improve a design. I would love to see any work you do. Try when using nupack analysis to use the temp settings of 27c min 5c Increment and 107c max it will provide you more detailed curve and match up with my observation.
Photo of Eli Fisker

Eli Fisker

  • 2273 Posts
  • 515 Reply Likes
Hi Joshua!

Here is the images from above, with same distance between the dots, as yours:

My design Mat with robotneck and tetraloops, without the nucleotide at position 4:



Here my design is as it were original, with the red nucleotide at position 4



Seems it becomes unstable a bit before, if the red nucleotide is added
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
looking at these graphs I would say that it was not an improvement. But to be honest I don't have a clue how to use these graphs to improve a design. I am using it like the hot cold game. I make a design then I compare it to Mat - Branches V2 melt profile if it look smiler I am happy if not I start over. If you are try to make any improvements to your design please document how you changed you design and how your graph has changed then if it gets selected to be synthesized maybe it will help improve our understanding of what these graphs mean. This is an image of the setting I use in Nupack just to help standardize our communication Good luck Eli Fisker
Photo of Eli Fisker

Eli Fisker

  • 2273 Posts
  • 515 Reply Likes
Mat made me take a extra look at your findings. As to what it could mean that the curves looked like this. As you (Joshua) suspected, I think your discovery can be used. It can be used to avoid making some of the ”worst” designs in the lab. It can also be used for sending a lab design through, to check if something might be wrong, before publishing. Or at least tell us if a design is stable at the temperature we want it to be stable at.

But first a bit about what I found out about this NUPACK analysis feature:

The way I read these curves, they are about at which temperature the RNA structure begins to fall apart – what amount of pairs that begins to become unstable/mispairing. Looking at the point where the curve begins to rise, the dot plot also begins to get messy. So I think a later rising curve is generally better, as it tells us that the design is stable at a bigger temperature range. The branches designs which have the lowest synthesis score, becomes unstable at a low temperature, sometimes even at 27 and 37 degree, which is viewable by changing temperature for the dot plot to the degree where the curves start to rise.

Here is what I discovered when I did following to Mat's Branches V2 (95%) and Aldo's Modulus (95%) design.

If I turn all GC-pairs in the opposite direction (save the one in the neck, as not to change anything here, for better or worse) the design becomes less stable, that is it start to mispair and fall apart at a lower temperature. One more proof for my rule about sameturning GC-pair in multiloops. :)

If I put in 4 energy gaps (by twisting some pairs in the middle of the strings) the curve raises a bit more quickly – that is the design becomes more unstable at a lower temperature. Actually underlines that energy gaps is not a good thing – and causes mispairing, which again causes temperature instability.

If I ad two more GC-pairs in the two stems connected to the middle loop, the curve rises more slowly than the originals. Which means adding GC-pairs ups the temperature stability. Which makes sense. I think it was Berex who told me that some RNA's living under extreme conditions, like heat, were made up of mostly GC-pairs.
And as we already know from the dot plot, adding more GC-pairs is not the same as it will synthesis well, even though it looks great in the tool. What the melting curve will look like, is just like the dot plot, connected to the numbers of GC-pairs.

If I ad just two GU-pairs – even at places where they would be best tolerated, the curve rises quicker at the middle. Towards a greater number of mispaired. Which is something also noticable in our usual dot plot. And unwanted.

Using the the common tetraloop with one red G, in general, gives a longer and more flat beginning. Meaning that it makes a design more temperature stable. This is propably why Vienna RNA-fold favors those, compared to all A's in the tetraloops. This is a general tendency for all the designs I checked, which is quite a few :)

General findings:

The curve should start somewhere around 0,45 fractions of bases unpaired. If it starts above 0,50, then something is wrong. And it should not rise too quickly.

The more overall linelike the ”curve”, the worse for you (like the robotdesigns with lowest synthesis score). Or with other words, the lower the synthesis score, the less curvelike structure in Nupack meltingpoint graph. The more temperature stable, the more caplike the Nupack analysis output becomes. The higher scoring the design, the more curvy (caplike is also possible) the outcome become.

The higher synthesis score the bots have, the more humanlike their curve.

This tool can be used to avoid making one of the lowest scorings of designs. It can be used to predict a low score. Actually when I took a look at my latest design (Mat with robotneck and tetraloops) in the Nupack analysis tool, it made me see, that I should propably not have put in that red nucleotide at position 4. Though it was not as big a change in the curve output, the RNA starts to mispair at a lower temperature. Though more change might be needed to improve my design, as the curve is not curvy enough in the right part of the graph. Which it should be to resemple the winning designs.

Now the synthesis result sort of confirms my suspisions. Synthesis score was 73% and I was modifying a 90% scoring design. Though it did worse than I hoped, I am happy for the lesson taught by my design and Weitzman's noticing of this funny caplike feature in the Nupack analysis tool.

Thereby this tool have kind of demonstrated it's value. It does many of the same things, that our usual dot plot tool already does, but this the dot plot tool could not have told me.
Photo of Eli Fisker

Eli Fisker

  • 2273 Posts
  • 515 Reply Likes
I have some correction to my post. The numbers for curvestart only holds for the good branches design. There seem to be an individual meltcurve for what is good design, for each lab puzzle.

Also I was a bit too quick to publish this, as the scores for the designround 5 went up. My design, which I thought scored 73% ended scoring 95%, so I actually improved it.

But why then did this red nucleotide change the meltcurve?

I now think what has happened with the melt curve of my design, is the same that happens with the colored view of the positional entropy. In Vienna RNA folds, when nucleotides are added in the hookarea, the overall impression of the entropy turns red. Which I have noticed doesn't nescessary means your design is good, but just covers up what mistakes there might be.

So nucleotides in hookarea might disturb the of the curve, so you can't use the tool to see if your design is good or bad.

Now I think the reason this red dot in the hook was allowed, was because it did not changed the energy level in the hook. Actually it might as well not have been there. I think it changed nothing for worse or better, except my ability to read the nupack tool correct.

My conclusion for now is: Don't use the nupack tool if you are planning to put nucleotides in the hookarea.
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
The Finger


The top 3 + Modified winner R1 overlayed black was 96
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
A % of pairs description would help in the labs went using a melt profile.
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
This is making me worried about my Terminator design look at the blue.
Photo of Eli Fisker

Eli Fisker

  • 2273 Posts
  • 515 Reply Likes
Hi Weitzman, I on to this. Yesterday Jee asked questions and I took a look in the old lab. It seems like that each puzzle have an individual curve for what is good or bad. Here is what Jee thinks: "It might have to do with how many stacks there are. The more stack there is, the harder for the RNA to break..I guess?"

So I think the most usefull comparison for your terminator, is the best branches puzzles. I voted on it as I'm looking forward to see how it does in lab.

So far I haven't drawn a conclusion about the different looking curves in the old lab yet. I like the way you put 4 puzzles onto one graph. Which program/tool do you use, to do that? Might have to learn that.
Photo of Joshua Weitzman

Joshua Weitzman

  • 93 Posts
  • 0 Reply Likes
I had to use Adobe Photoshop. Nupack don't have and Execel output I am worried about terminator design. I have a new theory I am working on how to compare the curves but I have to collect more data sets like the recent ones i publish. but as a preview I am guessing that if the curve moves slightly down and to the right of the targets curve then its an improvement. but it just move to the right it might not be good.

but to Jee question I think he is right the amount of stacks affects the shape of the curve plus how the stack are attracted to there correct pairs and how they have to repel the other stacks at the same time. Oh please call me Josh. Weitzman is a pain to write.