< User:ElNando888 | Blog
I was particularly impatient to see the results of Round 77, for a very simple reason: two of my lab proposals were running. So I will start by making an apology to the administrators of RMDB. Yes, the fool who's been (manually) hammering the server, reloading the http://rmdb.stanford.edu/repository/detail/ETERNA_R77_0000 page a few millions times every day for the past ten days, is me. Sorry about that.
New Frontier 2
http://eterna.cmu.edu/web/lab/3213788/ (moved to http://eterna.cmu.edu/web/lab/3376237/)
This lab was not much different than the first in the series, but I was expecting possible difficulties. In silico, it was more difficult to stabilize the shape, so I had imagined that the designs could display a greater propensity to misfold, specially taking into account that the knotting stem was locked with GC pairs exclusively.
Results and analysis
A rapid global inspection of the results shows that the pseudoknot simply formed in nearly all designs. I had submitted myself a couple attempts at preventing it, which totally failed in that regard. My own third submission was an attempt at "fooling" the system, creating basically a misfold which should have mimicked the SHAPE signature of a successful design. This one failed too, but I can't tell if it was caused by my lack of talent in designing for such targets, or if it was something else entirely...
Globally, this result is comforting. We can design various types of pseudoknots. It doesn't look like we are being limited to any special topology, so I guess that I will keep experimenting around those motifs, in an attempt to learn more about them. This said, we may have to wait a little before going further. Right now, I feel that the most interesting experiments in this area would be to have players design their pseudoknots with more freedom, so that we can test some limits. For instance, the first installment had a stack 11 bp long, areas of lengths 3, 0, and 10 unpaired bases, and a stem 6 bp long. The next natural step is to test various lengths of loops, keeping the stacks with the same lengths, so as to determine what is "mechanically" possible (in 3D), and what is clearly out of range. We can't do this yet in EteRNA, but I hear it will be soon possible. Stay tuned.
Dimer A & B
I can't hide that I'm very pleased with the results of this lab. Not because it worked as planned (actually, I think it didn't), but because it seems to give a lot of insights about the inner workings of Cloud Lab.
Results and analysis
The first direct observation is that there is indeed a good amount of protection signal in the 7-18 area in both labs. While this looks at first like a "success" (I had no idea if there would be any protection signal at all), the details seem to indicate that things didn't happen as expected.
For starters, it is extremely surprising that many protection signals are so strong. Dimerization is a function of concentration, and I had requested 80 slots so as to make the relative concentration of the pairing sequences something like 5%, since I was counting on 1600 sequences in the batch. Round 77 had actually 3378. In short, I was expecting a light shade of blue (if any), and it turns out that it should have been twice as light...
Roughly, the dimerizing segments can be divided in sub areas: a GC-poor region in the middle, flanked by 2 GC-richer regions. The results in both labs show that the protection signal is on average much weaker in the GC-poor region than in the flanking ones. This seems to suggest that these dimerizing segments may have paired with their counterparts, but it is very unlikely that it was the only thing they paired with. And if we assume a low baseline for the number of fully paired dimers, then what can explain the stronger signal on the GC-rich regions?
The first possibility that comes to mind is a misfold. These segments meant for dimerizing would have paired with something somewhere else in the same molecule. But the results do not support the idea that the candidates had trouble folding properly in the areas that were not meant to dimerize.
Another misfold hypothesis is being discussed in the talk page, but (so far) this one too, seems quite unlikely.
Finally, it seems conceivable that these regions simply paired with any compatible segment of any other design in some other lab in Cloud Lab. And it so happens that there's an interesting fact about this specific round: the global error rate is higher than usual... Oops (?)
Among the Dimer B submissions, there is a highly interesting design submitted by Stlnegril9, namely Primer Dimer B. By the looks of the dotplot, this misfold was 100% intentional, and I'm very glad that he decided to submit this, because it shows that intramolecular interactions prevent intermolecular ones. The bases 7 to 10 in this design have a clearly reactive signal, meaning that the dimer duplexes practically never formed.
Conclusions (so far)
- Long locked unpaired GC-rich sequences (like in this Dimer A & B labs) are probably not a good idea in the context of Cloud Lab
- Xmas trees, on top of clearly failing consistently, may be detrimental for the Cloud Lab as a whole.