User:ElNando888/Blog/Models 2

From Eterna Wiki




In an earlier blog post, I presented the 3 models that are available in the current implementation of ViennaRNA (at the time of this writing, version 2.1.6). 

  • Turner 1999
  • Turner 2004
  • Andronescu 2007

On a single test of prediction, it appeared that Andronescu 2007 was performing better. But of course, that was only one test...


The AU closing base-pair case


This time, I was interested in evaluating the estimated likelihood of the formation of an AU pair closing a tetraloop.

With a very short sequence, I used the ViennaRNA online folding tool to create this table

GCGUACACACGC Turner 1999 Turner 2004 Andronescu 2007
Blog model2 HL1a.png -2.70 -2.70 -2.85
Blog model2 HL1b.png -2.30 -1.50 -0.94
  +0.4 +1.2 +1.91
MFE Probability 63.84% 82.36% 85.23%


What does this all mean? To make a long story short:

  • all models think that the UA pair will form
  • Turner 1999 thinks that this pair has about a 35% chance of opening up
  • Turner 2004 thinks that this is less likely than that, only around 17%
  • Andronescu 2007 goes even further and evaluates that chance at about 15%


A lab result

Blog models2 SHAPE.png


This result comes from the most recent round, in the Late Bulge Control lab. If I'm not mistaken, this is far from being the first time that one sees such kind of SHAPE signal for AU closing base-pairs.

Turns out that here, the Andronescu 2007 model is the wrongest of all, and the best prediction comes from Turner 1999


A 3D re-creation

Blog model2 ROSIE.png

This is a model I created with the help of "Rhiju's baby", RNA denovo, on the ROSIE server.

The job went apparently quite well, achieving atomic accuracy. Notice that the UA pair is NOT a canonical one. If I'm not mistaken, this is a Reversed Hoogsteen AU. So locally, this should be regarded as a "boost" for a hexaloop.


Concluding remarks

(throwing ideas, needs rephrasing)

(point made, all models are imperfect, unclear which one is "better")

(the last Turner publication of such a set of parameters is 10 years old, they waited only 5 before, will there be a new one any time soon?)

(is there a chance that we could improve these parameters and create a new one?)